* [PATCH v3 0/3] Allow to build indexes for vinyl spaces
@ 2018-06-07 10:56 Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 1/3] vinyl: do not yield on dump completion Vladimir Davydov
` (3 more replies)
0 siblings, 4 replies; 5+ messages in thread
From: Vladimir Davydov @ 2018-06-07 10:56 UTC (permalink / raw)
To: kostja; +Cc: tarantool-patches
This patch set implements the ability to build secondary indexes for
non-empty vinyl spaces. For implementation details, see patch 3.
https://github.com/tarantool/tarantool/issues/1653
https://github.com/tarantool/tarantool/commits/vy-allow-to-build-secondary-indexes
Changes in v3:
- Remove merged patches from the patch set.
- Rebase tests on top of the latest 1.10.
- When building a new index, use original tuple LSNs instead
of max LSN to the time.
Changes in v2:
- Split the patch into logical parts.
- Address comments by Konstantin and Vladislav (refactoring mostly).
v1: https://www.freelists.org/post/tarantool-patches/PATCH-vinyl-allow-to-build-secondary-index-for-nonempty-space
v2: https://www.freelists.org/post/tarantool-patches/PATCH-v2-08-Allow-to-build-indexes-for-vinyl-spaces
Vladimir Davydov (3):
vinyl: do not yield on dump completion
vinyl: relax limitation imposed on run min/max lsn
vinyl: allow to build secondary index for non-empty space
src/box/vinyl.c | 450 +++++++++++++++++++++++---
src/box/vy_log.c | 2 +-
src/box/vy_log.h | 4 +-
src/box/vy_lsm.c | 14 +-
src/box/vy_lsm.h | 2 +-
src/box/vy_quota.h | 10 +
src/box/vy_read_iterator.c | 13 -
src/box/vy_scheduler.c | 71 +++--
src/box/vy_scheduler.h | 7 +
test/box/alter.result | 501 -----------------------------
test/box/alter.test.lua | 159 ---------
test/engine/ddl.result | 672 +++++++++++++++++++++++++++++++++++++++
test/engine/ddl.test.lua | 221 +++++++++++++
test/vinyl/ddl.result | 439 ++++++++++++++-----------
test/vinyl/ddl.test.lua | 231 +++++++++-----
test/vinyl/errinj.result | 312 ++++++++++++++++++
test/vinyl/errinj.test.lua | 134 ++++++++
test/vinyl/errinj_gc.result | 78 ++++-
test/vinyl/errinj_gc.test.lua | 38 ++-
test/vinyl/errinj_vylog.result | 57 ++++
test/vinyl/errinj_vylog.test.lua | 29 ++
test/vinyl/gh.result | 2 +-
22 files changed, 2396 insertions(+), 1050 deletions(-)
--
2.11.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v3 1/3] vinyl: do not yield on dump completion
2018-06-07 10:56 [PATCH v3 0/3] Allow to build indexes for vinyl spaces Vladimir Davydov
@ 2018-06-07 10:56 ` Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 2/3] vinyl: relax limitation imposed on run min/max lsn Vladimir Davydov
` (2 subsequent siblings)
3 siblings, 0 replies; 5+ messages in thread
From: Vladimir Davydov @ 2018-06-07 10:56 UTC (permalink / raw)
To: kostja; +Cc: tarantool-patches
The fact that we may yield after we added a new slice created by dump,
but before we removed the dumped in-memory index from the LSM tree
complicates read iterator logic, as it has to detect such a case and
filter out sources that contain duplicates. This logic relies on the
fact that no two slices of the same range intersect by LSN. For the
sake of ALTER we have to relax this limitation, as statements inserted
during index build can have arbitrary (not monotonically growing) LSNs,
so the no-LSN-intersection property won't be fulfilled for whole slices,
only for individual keys. Since there shouldn't be more than 1000 ranges
in the same LSM tree, yielding doesn't make much sense as iteration over
the whole range tree should be pretty fast. Besides, dump isn't done
frequently. That said, let's remove yielding altogether.
Needed for #1653
---
src/box/vy_read_iterator.c | 13 -------------
src/box/vy_scheduler.c | 39 ++++++---------------------------------
2 files changed, 6 insertions(+), 46 deletions(-)
diff --git a/src/box/vy_read_iterator.c b/src/box/vy_read_iterator.c
index 61c3b683..d8d66229 100644
--- a/src/box/vy_read_iterator.c
+++ b/src/box/vy_read_iterator.c
@@ -635,19 +635,6 @@ vy_read_iterator_add_disk(struct vy_read_iterator *itr)
* format in vy_mem.
*/
rlist_foreach_entry(slice, &itr->curr_range->slices, in_range) {
- /*
- * vy_task_dump_complete() may yield after adding
- * a new run slice to a range and before removing
- * dumped in-memory trees. We must not add both
- * the slice and the trees in this case, because
- * the read iterator can't deal with duplicates.
- * Since lsm->dump_lsn is bumped after deletion
- * of dumped in-memory trees, we can filter out
- * the run slice containing duplicates by LSN.
- */
- if (slice->run->info.min_lsn > lsm->dump_lsn)
- continue;
- assert(slice->run->info.max_lsn <= lsm->dump_lsn);
struct vy_read_src *sub_src = vy_read_iterator_add_src(itr);
vy_run_iterator_open(&sub_src->run_iterator,
&lsm->stat.disk.iterator, slice,
diff --git a/src/box/vy_scheduler.c b/src/box/vy_scheduler.c
index 3631f644..f4746d68 100644
--- a/src/box/vy_scheduler.c
+++ b/src/box/vy_scheduler.c
@@ -57,16 +57,6 @@
#include "trivia/util.h"
#include "tt_pthread.h"
-/**
- * Yield after iterating over this many objects (e.g. ranges).
- * Yield more often in debug mode.
- */
-#if defined(NDEBUG)
-enum { VY_YIELD_LOOPS = 128 };
-#else
-enum { VY_YIELD_LOOPS = 2 };
-#endif
-
/* Min and max values for vy_scheduler::timeout. */
#define VY_SCHEDULER_TIMEOUT_MIN 1
#define VY_SCHEDULER_TIMEOUT_MAX 60
@@ -712,7 +702,7 @@ vy_task_dump_complete(struct vy_scheduler *scheduler, struct vy_task *task)
struct vy_slice **new_slices, *slice;
struct vy_range *range, *begin_range, *end_range;
struct tuple *min_key, *max_key;
- int i, loops = 0;
+ int i;
assert(lsm->is_dumping);
@@ -777,12 +767,6 @@ vy_task_dump_complete(struct vy_scheduler *scheduler, struct vy_task *task)
assert(i < lsm->range_count);
new_slices[i] = slice;
- /*
- * It's OK to yield here for the range tree can only
- * be changed from the scheduler fiber.
- */
- if (++loops % VY_YIELD_LOOPS == 0)
- fiber_sleep(0);
}
/*
@@ -797,9 +781,6 @@ vy_task_dump_complete(struct vy_scheduler *scheduler, struct vy_task *task)
vy_log_insert_slice(range->id, new_run->id, slice->id,
tuple_data_or_null(slice->begin),
tuple_data_or_null(slice->end));
-
- if (++loops % VY_YIELD_LOOPS == 0)
- fiber_sleep(0); /* see comment above */
}
vy_log_dump_lsm(lsm->id, dump_lsn);
if (vy_log_tx_commit() < 0)
@@ -816,6 +797,11 @@ vy_task_dump_complete(struct vy_scheduler *scheduler, struct vy_task *task)
/*
* Add new slices to ranges.
+ *
+ * Note, we must not yield after this point, because if we
+ * do, a concurrent read iterator may see an inconsistent
+ * LSM tree state, when the same statement is present twice,
+ * in memory and on disk.
*/
for (range = begin_range, i = 0; range != end_range;
range = vy_range_tree_next(lsm->tree, range), i++) {
@@ -829,17 +815,6 @@ vy_task_dump_complete(struct vy_scheduler *scheduler, struct vy_task *task)
vy_range_heap_update(&lsm->range_heap,
&range->heap_node);
range->version++;
- /*
- * If we yield here, a concurrent fiber will see
- * a range with a run slice containing statements
- * present in the in-memory indexes of the LSM tree.
- * This is OK, because read iterator won't use the
- * new run slice until lsm->dump_lsn is bumped,
- * which is only done after in-memory trees are
- * removed (see vy_read_iterator_add_disk()).
- */
- if (++loops % VY_YIELD_LOOPS == 0)
- fiber_sleep(0);
}
free(new_slices);
@@ -878,8 +853,6 @@ fail_free_slices:
slice = new_slices[i];
if (slice != NULL)
vy_slice_delete(slice);
- if (++loops % VY_YIELD_LOOPS == 0)
- fiber_sleep(0);
}
free(new_slices);
fail:
--
2.11.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v3 2/3] vinyl: relax limitation imposed on run min/max lsn
2018-06-07 10:56 [PATCH v3 0/3] Allow to build indexes for vinyl spaces Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 1/3] vinyl: do not yield on dump completion Vladimir Davydov
@ 2018-06-07 10:56 ` Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 3/3] vinyl: allow to build secondary index for non-empty space Vladimir Davydov
2018-06-07 14:02 ` [PATCH v3 0/3] Allow to build indexes for vinyl spaces Konstantin Osipov
3 siblings, 0 replies; 5+ messages in thread
From: Vladimir Davydov @ 2018-06-07 10:56 UTC (permalink / raw)
To: kostja; +Cc: tarantool-patches
Currently, we assume that no two runs of the same range intersect by
LSN. This holds, because LSNs grow strictly monotonically, and no
transaction may be split between two runs (as we pin each affected
vy_mem until the transaction is complete). We ensure this with an
assertion in vy_task_dump_complete.
However, tuples inserted during index build will have arbitrary (not
monotonically growing) LSNs. This is OK as for each particular key, two
statements will still have different LSNs, but this may break the
assertion in vy_task_dump_complete in case dump occurs while build is in
progress.
To avoid that, let's relax this limitation and assume that a dumped run
may intersect by LSN with runs dumped before. Moreover, let's assume
that it may have max LSN less than the max LSN stored on disk so that
we should update vy_lsm::dump_lsn only if the dumped run has newer data.
Needed for #1653
---
src/box/vy_log.c | 2 +-
src/box/vy_log.h | 4 ++--
src/box/vy_lsm.h | 2 +-
src/box/vy_scheduler.c | 3 +--
4 files changed, 5 insertions(+), 6 deletions(-)
diff --git a/src/box/vy_log.c b/src/box/vy_log.c
index 8da457a6..6556dd37 100644
--- a/src/box/vy_log.c
+++ b/src/box/vy_log.c
@@ -1485,7 +1485,7 @@ vy_recovery_dump_lsm(struct vy_recovery *recovery,
(long long)id));
return -1;
}
- lsm->dump_lsn = dump_lsn;
+ lsm->dump_lsn = MAX(lsm->dump_lsn, dump_lsn);
return 0;
}
diff --git a/src/box/vy_log.h b/src/box/vy_log.h
index 442563f0..0a216de8 100644
--- a/src/box/vy_log.h
+++ b/src/box/vy_log.h
@@ -136,7 +136,7 @@ enum vy_log_record_type {
*/
VY_LOG_DELETE_SLICE = 9,
/**
- * Update LSN of the last LSM tree dump.
+ * Log LSM tree dump. Used to update max LSN stored on disk.
* Requires vy_log_record::lsm_id, dump_lsn.
*/
VY_LOG_DUMP_LSM = 10,
@@ -303,7 +303,7 @@ struct vy_lsm_recovery_info {
* if the tree is still active.
*/
int64_t drop_lsn;
- /** LSN of the last LSM tree dump. */
+ /** Max LSN stored on disk. */
int64_t dump_lsn;
/**
* List of all ranges in the LSM tree, linked by
diff --git a/src/box/vy_lsm.h b/src/box/vy_lsm.h
index 2e99e4d0..90ccb534 100644
--- a/src/box/vy_lsm.h
+++ b/src/box/vy_lsm.h
@@ -253,7 +253,7 @@ struct vy_lsm {
*/
uint32_t range_tree_version;
/**
- * LSN of the last dump or -1 if the LSM tree has not
+ * Max LSN stored on disk or -1 if the LSM tree has not
* been dumped yet.
*/
int64_t dump_lsn;
diff --git a/src/box/vy_scheduler.c b/src/box/vy_scheduler.c
index f4746d68..e3f2e223 100644
--- a/src/box/vy_scheduler.c
+++ b/src/box/vy_scheduler.c
@@ -721,7 +721,6 @@ vy_task_dump_complete(struct vy_scheduler *scheduler, struct vy_task *task)
goto delete_mems;
}
- assert(new_run->info.min_lsn > lsm->dump_lsn);
assert(new_run->info.max_lsn <= dump_lsn);
/*
@@ -828,7 +827,7 @@ delete_mems:
vy_stmt_counter_add(&lsm->stat.disk.dump.in, &mem->count);
vy_lsm_delete_mem(lsm, mem);
}
- lsm->dump_lsn = dump_lsn;
+ lsm->dump_lsn = MAX(lsm->dump_lsn, dump_lsn);
lsm->stat.disk.dump.count++;
/* The iterator has been cleaned up in a worker thread. */
--
2.11.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v3 3/3] vinyl: allow to build secondary index for non-empty space
2018-06-07 10:56 [PATCH v3 0/3] Allow to build indexes for vinyl spaces Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 1/3] vinyl: do not yield on dump completion Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 2/3] vinyl: relax limitation imposed on run min/max lsn Vladimir Davydov
@ 2018-06-07 10:56 ` Vladimir Davydov
2018-06-07 14:02 ` [PATCH v3 0/3] Allow to build indexes for vinyl spaces Konstantin Osipov
3 siblings, 0 replies; 5+ messages in thread
From: Vladimir Davydov @ 2018-06-07 10:56 UTC (permalink / raw)
To: kostja; +Cc: tarantool-patches
This patch implements space_vtab::build_index callback for vinyl spaces.
Now instead of returning an error in case the space is not empty, the
callback will actually try to build a new index. The build procedure
consists of four steps:
1. Prepare the LSM tree for building. This implies writing a special
record to vylog, VY_LOG_PREPARE_LSM, and adding the new index to
the vinyl scheduler so that it can be dumped during build. We need
to log the new LSM tree so that we can keep track of run files
created for it during build and remove them if build procedure
fails.
2. Inserting tuples stored in the space into the new LSM tree. Since
there may concurrent DML requests, we install a trigger to forward
them to the new index.
3. Dumping the index to disk so that we don't have to rebuild it after
recovery.
4. Committing the new LSM tree in vylog (VY_LOG_CREATE_LSM).
Steps 1-3 are done from the space_vtab::build_index callback while
step 4 is done after WAL write, from index_vtab::commit_create.
While step 3 is being performed, new DML requests may be executed for
the altered space. Those requests will be reflected in the new index
thanks to the on_replace trigger, however they won't be recovered during
WAL recovery as they will appear in WAL before the ALTER record that
created the index. To recover them, we replay all statements stored in
the primary key's memory level when replaying the ALTER record during
WAL recovery.
Closes #1653
---
src/box/vinyl.c | 450 +++++++++++++++++++++++---
src/box/vy_lsm.c | 14 +-
src/box/vy_quota.h | 10 +
src/box/vy_scheduler.c | 29 ++
src/box/vy_scheduler.h | 7 +
test/box/alter.result | 501 -----------------------------
test/box/alter.test.lua | 159 ---------
test/engine/ddl.result | 672 +++++++++++++++++++++++++++++++++++++++
test/engine/ddl.test.lua | 221 +++++++++++++
test/vinyl/ddl.result | 439 ++++++++++++++-----------
test/vinyl/ddl.test.lua | 231 +++++++++-----
test/vinyl/errinj.result | 312 ++++++++++++++++++
test/vinyl/errinj.test.lua | 134 ++++++++
test/vinyl/errinj_gc.result | 78 ++++-
test/vinyl/errinj_gc.test.lua | 38 ++-
test/vinyl/errinj_vylog.result | 57 ++++
test/vinyl/errinj_vylog.test.lua | 29 ++
test/vinyl/gh.result | 2 +-
18 files changed, 2385 insertions(+), 998 deletions(-)
diff --git a/src/box/vinyl.c b/src/box/vinyl.c
index fd5aa7a5..d2e3da7e 100644
--- a/src/box/vinyl.c
+++ b/src/box/vinyl.c
@@ -1164,52 +1164,6 @@ vinyl_space_drop_primary_key(struct space *space)
(void)space;
}
-static int
-vinyl_space_build_index(struct space *src_space, struct index *new_index,
- struct tuple_format *new_format)
-{
- (void)new_format;
-
- struct vy_env *env = vy_env(src_space->engine);
- struct vy_lsm *pk = vy_lsm(src_space->index[0]);
-
- /*
- * During local recovery we are loading existing indexes
- * from disk, not building new ones.
- */
- if (env->status != VINYL_INITIAL_RECOVERY_LOCAL &&
- env->status != VINYL_FINAL_RECOVERY_LOCAL) {
- if (pk->stat.disk.count.rows != 0 ||
- pk->stat.memory.count.rows != 0) {
- diag_set(ClientError, ER_UNSUPPORTED, "Vinyl",
- "building an index for a non-empty space");
- return -1;
- }
- }
-
- /*
- * Unlike Memtx, Vinyl does not need building of a secondary index.
- * This is true because of two things:
- * 1) Vinyl does not support alter of non-empty spaces
- * 2) During recovery a Vinyl index already has all needed data on disk.
- * And there are 3 cases:
- * I. The secondary index is added in snapshot. Then Vinyl was
- * snapshotted too and all necessary for that moment data is on disk.
- * II. The secondary index is added in WAL. That means that vinyl
- * space had no data at that point and had nothing to build. The
- * index actually could contain recovered data, but it will handle it
- * by itself during WAL recovery.
- * III. Vinyl is online. The space is definitely empty and there's
- * nothing to build.
- */
- if (vinyl_index_open(new_index) != 0)
- return -1;
-
- /* Set pointer to the primary key for the new index. */
- vy_lsm_update_pk(vy_lsm(new_index), pk);
- return 0;
-}
-
static size_t
vinyl_space_bsize(struct space *space)
{
@@ -4092,6 +4046,410 @@ vinyl_index_get(struct index *index, const char *key,
/*** }}} Cursor */
+/* {{{ Index build */
+
+/** Argument passed to vy_build_on_replace(). */
+struct vy_build_ctx {
+ /** Vinyl environment. */
+ struct vy_env *env;
+ /** LSM tree under construction. */
+ struct vy_lsm *lsm;
+ /** Format to check new tuples against. */
+ struct tuple_format *format;
+ /**
+ * Names of the altered space and the new index.
+ * Used for error reporting.
+ */
+ const char *space_name;
+ const char *index_name;
+ /** Set in case a build error occurred. */
+ bool is_failed;
+ /** Container for storing errors. */
+ struct diag diag;
+};
+
+/**
+ * This is an on_replace trigger callback that forwards DML requests
+ * to the index that is currently being built.
+ */
+static void
+vy_build_on_replace(struct trigger *trigger, void *event)
+{
+ struct txn *txn = event;
+ struct txn_stmt *stmt = txn_current_stmt(txn);
+ struct vy_build_ctx *ctx = trigger->data;
+ struct vy_tx *tx = txn->engine_tx;
+ struct tuple_format *format = ctx->format;
+ struct vy_lsm *lsm = ctx->lsm;
+
+ if (ctx->is_failed)
+ return; /* already failed, nothing to do */
+
+ /* Check new tuples for conformity to the new format. */
+ if (stmt->new_tuple != NULL &&
+ tuple_validate(format, stmt->new_tuple) != 0)
+ goto err;
+
+ /* Check key uniqueness if necessary. */
+ if (stmt->new_tuple != NULL &&
+ vy_check_is_unique_secondary(ctx->env, tx, vy_tx_read_view(tx),
+ ctx->space_name, ctx->index_name,
+ lsm, stmt->new_tuple) != 0)
+ goto err;
+
+ /* Forward the statement to the new LSM tree. */
+ if (stmt->old_tuple != NULL) {
+ struct tuple *delete = vy_stmt_new_surrogate_delete(format,
+ stmt->old_tuple);
+ if (delete == NULL)
+ goto err;
+ int rc = vy_tx_set(tx, lsm, delete);
+ tuple_unref(delete);
+ if (rc != 0)
+ goto err;
+ }
+ if (stmt->new_tuple != NULL) {
+ uint32_t data_len;
+ const char *data = tuple_data_range(stmt->new_tuple, &data_len);
+ struct tuple *insert = vy_stmt_new_insert(format, data,
+ data + data_len);
+ if (insert == NULL)
+ goto err;
+ int rc = vy_tx_set(tx, lsm, insert);
+ tuple_unref(insert);
+ if (rc != 0)
+ goto err;
+ }
+ return;
+err:
+ ctx->is_failed = true;
+ diag_move(diag_get(), &ctx->diag);
+}
+
+/**
+ * Insert a single statement into the LSM tree that is currently
+ * being built.
+ */
+static int
+vy_build_insert_stmt(struct vy_lsm *lsm, struct vy_mem *mem,
+ const struct tuple *stmt, int64_t lsn)
+{
+ const struct tuple *region_stmt = vy_stmt_dup_lsregion(stmt,
+ &mem->env->allocator, mem->generation);
+ if (region_stmt == NULL)
+ return -1;
+ vy_stmt_set_lsn((struct tuple *)region_stmt, lsn);
+ if (vy_mem_insert(mem, region_stmt) != 0)
+ return -1;
+ vy_mem_commit_stmt(mem, region_stmt);
+ vy_stmt_counter_acct_tuple(&lsm->stat.memory.count, region_stmt);
+ return 0;
+}
+
+/**
+ * Insert a tuple fetched from the space into the LSM tree that
+ * is currently being built.
+ */
+static int
+vy_build_insert_tuple(struct vy_env *env, struct vy_lsm *lsm,
+ const char *space_name, const char *index_name,
+ struct tuple_format *new_format, struct tuple *tuple)
+{
+ int rc;
+ struct vy_mem *mem = lsm->mem;
+ int64_t lsn = vy_stmt_lsn(tuple);
+
+ /* Check the tuple against the new space format. */
+ if (tuple_validate(new_format, tuple) != 0)
+ return -1;
+
+ /*
+ * Check unique constraint if necessary.
+ *
+ * Note, this operation may yield, which opens a time
+ * window for a concurrent fiber to insert a newer tuple
+ * version. It's OK - we won't overwrite it, because the
+ * LSN we use is less. However, we do need to make sure
+ * we insert the tuple into the in-memory index that was
+ * active before the yield, otherwise we might break the
+ * invariant according to which newer in-memory indexes
+ * store statements with greater LSNs. So we pin the
+ * in-memory index that is active now and insert the tuple
+ * into it after the yield.
+ */
+ vy_mem_pin(mem);
+ rc = vy_check_is_unique_secondary(env, NULL,
+ &env->xm->p_committed_read_view,
+ space_name, index_name, lsm, tuple);
+ vy_mem_unpin(mem);
+ if (rc != 0)
+ return -1;
+
+ /* Reallocate the new tuple using the new space format. */
+ uint32_t data_len;
+ const char *data = tuple_data_range(tuple, &data_len);
+ struct tuple *stmt = vy_stmt_new_replace(new_format, data,
+ data + data_len);
+ if (stmt == NULL)
+ return -1;
+
+ /* Insert the new tuple into the in-memory index. */
+ size_t mem_used_before = lsregion_used(&env->mem_env.allocator);
+ rc = vy_build_insert_stmt(lsm, mem, stmt, lsn);
+ tuple_unref(stmt);
+
+ /* Consume memory quota. Throttle if it is exceeded. */
+ size_t mem_used_after = lsregion_used(&env->mem_env.allocator);
+ assert(mem_used_after >= mem_used_before);
+ vy_quota_force_use(&env->quota, mem_used_after - mem_used_before);
+ vy_quota_wait(&env->quota);
+ return rc;
+}
+
+/**
+ * Recover a single statement that was inserted into the space
+ * while the newly built index was dumped to disk.
+ */
+static int
+vy_build_recover_stmt(struct vy_lsm *lsm, struct vy_lsm *pk,
+ const struct tuple *mem_stmt)
+{
+ int64_t lsn = vy_stmt_lsn(mem_stmt);
+ if (lsn <= lsm->dump_lsn)
+ return 0; /* statement was dumped, nothing to do */
+
+ /* Lookup the tuple that was affected by this statement. */
+ const struct vy_read_view rv = { .vlsn = lsn - 1 };
+ const struct vy_read_view *p_rv = &rv;
+ struct tuple *old_tuple;
+ if (vy_point_lookup(pk, NULL, &p_rv, (struct tuple *)mem_stmt,
+ &old_tuple) != 0)
+ return -1;
+ /*
+ * Create DELETE + INSERT statements corresponding to
+ * the given statement in the secondary index.
+ */
+ struct tuple *delete = NULL;
+ struct tuple *insert = NULL;
+ if (old_tuple != NULL) {
+ delete = vy_stmt_new_surrogate_delete(lsm->mem_format,
+ old_tuple);
+ if (delete == NULL)
+ return -1;
+ }
+ enum iproto_type type = vy_stmt_type(mem_stmt);
+ if (type == IPROTO_REPLACE || type == IPROTO_INSERT) {
+ uint32_t data_len;
+ const char *data = tuple_data_range(mem_stmt, &data_len);
+ insert = vy_stmt_new_insert(lsm->mem_format,
+ data, data + data_len);
+ if (insert == NULL)
+ return -1;
+ } else if (type == IPROTO_UPSERT) {
+ struct tuple *new_tuple = vy_apply_upsert(mem_stmt, old_tuple,
+ pk->cmp_def, pk->mem_format, true);
+ if (new_tuple == NULL)
+ return -1;
+ uint32_t data_len;
+ const char *data = tuple_data_range(new_tuple, &data_len);
+ insert = vy_stmt_new_insert(lsm->mem_format,
+ data, data + data_len);
+ tuple_unref(new_tuple);
+ if (insert == NULL)
+ return -1;
+ }
+
+ /* Insert DELETE + INSERT into the LSM tree. */
+ if (delete != NULL) {
+ int rc = vy_build_insert_stmt(lsm, lsm->mem, delete, lsn);
+ tuple_unref(delete);
+ if (rc != 0)
+ return -1;
+ }
+ if (insert != NULL) {
+ int rc = vy_build_insert_stmt(lsm, lsm->mem, insert, lsn);
+ tuple_unref(insert);
+ if (rc != 0)
+ return -1;
+ }
+ return 0;
+}
+
+/**
+ * Recover all statements stored in the given in-memory index
+ * that were inserted into the space while the newly built index
+ * was dumped to disk.
+ */
+static int
+vy_build_recover_mem(struct vy_lsm *lsm, struct vy_lsm *pk, struct vy_mem *mem)
+{
+ /*
+ * Recover statements starting from the oldest one.
+ * Key order doesn't matter so we simply iterate over
+ * the in-memory index in reverse order.
+ */
+ struct vy_mem_tree_iterator itr;
+ itr = vy_mem_tree_iterator_last(&mem->tree);
+ while (!vy_mem_tree_iterator_is_invalid(&itr)) {
+ const struct tuple *mem_stmt;
+ mem_stmt = *vy_mem_tree_iterator_get_elem(&mem->tree, &itr);
+ if (vy_build_recover_stmt(lsm, pk, mem_stmt) != 0)
+ return -1;
+ vy_mem_tree_iterator_prev(&mem->tree, &itr);
+ }
+ return 0;
+}
+
+/**
+ * Recover the memory level of a newly built index.
+ *
+ * During the final dump of a newly built index, new statements may
+ * be inserted into the space. If those statements are not dumped to
+ * disk before restart, they won't be recovered from WAL, because at
+ * the time they were generated the new index didn't exist. In order
+ * to recover them, we replay all statements stored in the memory
+ * level of the primary index.
+ */
+static int
+vy_build_recover(struct vy_env *env, struct vy_lsm *lsm, struct vy_lsm *pk)
+{
+ int rc = 0;
+ struct vy_mem *mem;
+ size_t mem_used_before, mem_used_after;
+
+ mem_used_before = lsregion_used(&env->mem_env.allocator);
+ rlist_foreach_entry_reverse(mem, &pk->sealed, in_sealed) {
+ rc = vy_build_recover_mem(lsm, pk, mem);
+ if (rc != 0)
+ break;
+ }
+ if (rc == 0)
+ rc = vy_build_recover_mem(lsm, pk, pk->mem);
+
+ mem_used_after = lsregion_used(&env->mem_env.allocator);
+ assert(mem_used_after >= mem_used_before);
+ vy_quota_force_use(&env->quota, mem_used_after - mem_used_before);
+ return rc;
+}
+
+static int
+vinyl_space_build_index(struct space *src_space, struct index *new_index,
+ struct tuple_format *new_format)
+{
+ struct vy_env *env = vy_env(src_space->engine);
+ struct vy_lsm *pk = vy_lsm(src_space->index[0]);
+ bool is_empty = (pk->stat.disk.count.rows == 0 &&
+ pk->stat.memory.count.rows == 0);
+
+ if (new_index->def->iid == 0 && !is_empty) {
+ diag_set(ClientError, ER_UNSUPPORTED, "Vinyl",
+ "rebuilding the primary index of a non-empty space");
+ return -1;
+ }
+
+ if (vinyl_index_open(new_index) != 0)
+ return -1;
+
+ /* Set pointer to the primary key for the new index. */
+ struct vy_lsm *new_lsm = vy_lsm(new_index);
+ vy_lsm_update_pk(new_lsm, pk);
+
+ if (env->status == VINYL_INITIAL_RECOVERY_LOCAL ||
+ env->status == VINYL_FINAL_RECOVERY_LOCAL)
+ return vy_build_recover(env, new_lsm, pk);
+
+ if (is_empty)
+ return 0;
+
+ /*
+ * Iterate over all tuples stored in the space and insert
+ * each of them into the new LSM tree. Since read iterator
+ * may yield, we install an on_replace trigger to forward
+ * DML requests issued during the build.
+ */
+ struct tuple *key = vy_stmt_new_select(pk->env->key_format, NULL, 0);
+ if (key == NULL)
+ return -1;
+
+ struct trigger on_replace;
+ struct vy_build_ctx ctx;
+ ctx.env = env;
+ ctx.lsm = new_lsm;
+ ctx.format = new_format;
+ ctx.space_name = space_name(src_space);
+ ctx.index_name = new_index->def->name;
+ ctx.is_failed = false;
+ diag_create(&ctx.diag);
+ trigger_create(&on_replace, vy_build_on_replace, &ctx, NULL);
+ trigger_add(&src_space->on_replace, &on_replace);
+
+ struct vy_read_iterator itr;
+ vy_read_iterator_open(&itr, pk, NULL, ITER_ALL, key,
+ &env->xm->p_committed_read_view);
+ int rc;
+ int loops = 0;
+ struct tuple *tuple;
+ int64_t build_lsn = env->xm->lsn;
+ while ((rc = vy_read_iterator_next(&itr, &tuple)) == 0) {
+ if (tuple == NULL)
+ break;
+ /*
+ * Insert the tuple into the new index unless it
+ * was inserted into the space after we started
+ * building the new index - in the latter case
+ * the new tuple has already been inserted by the
+ * on_replace trigger.
+ *
+ * Note, yield is not allowed between reading the
+ * tuple from the primary index and inserting it
+ * into the new index. If we yielded, the tuple
+ * could be overwritten by a concurrent transaction,
+ * in which case we would insert an outdated tuple.
+ */
+ if (vy_stmt_lsn(tuple) <= build_lsn) {
+ rc = vy_build_insert_tuple(env, new_lsm,
+ space_name(src_space),
+ new_index->def->name,
+ new_format, tuple);
+ if (rc != 0)
+ break;
+ }
+ /*
+ * Read iterator yields only when it reads runs.
+ * Yield periodically in order not to stall the
+ * tx thread in case there are a lot of tuples in
+ * mems or cache.
+ */
+ if (++loops % VY_YIELD_LOOPS == 0)
+ fiber_sleep(0);
+ if (ctx.is_failed) {
+ diag_move(&ctx.diag, diag_get());
+ rc = -1;
+ break;
+ }
+ }
+ vy_read_iterator_close(&itr);
+ tuple_unref(key);
+
+ /*
+ * Dump the new index upon build completion so that we don't
+ * have to rebuild it on recovery.
+ */
+ if (rc == 0)
+ rc = vy_scheduler_dump(&env->scheduler);
+
+ if (rc == 0 && ctx.is_failed) {
+ diag_move(&ctx.diag, diag_get());
+ rc = -1;
+ }
+
+ diag_destroy(&ctx.diag);
+ trigger_clear(&on_replace);
+ return rc;
+}
+
+/* }}} Index build */
+
static const struct engine_vtab vinyl_engine_vtab = {
/* .shutdown = */ vinyl_engine_shutdown,
/* .create_space = */ vinyl_engine_create_space,
diff --git a/src/box/vy_lsm.c b/src/box/vy_lsm.c
index fd352c8a..5ac5548b 100644
--- a/src/box/vy_lsm.c
+++ b/src/box/vy_lsm.c
@@ -773,11 +773,20 @@ int
vy_lsm_set(struct vy_lsm *lsm, struct vy_mem *mem,
const struct tuple *stmt, const struct tuple **region_stmt)
{
+ uint32_t format_id = stmt->format_id;
+
assert(vy_stmt_is_refable(stmt));
assert(*region_stmt == NULL || !vy_stmt_is_refable(*region_stmt));
- /* Allocate region_stmt on demand. */
- if (*region_stmt == NULL) {
+ /*
+ * Allocate region_stmt on demand.
+ *
+ * Also, reallocate region_stmt if it uses a different tuple
+ * format. This may happen during ALTER, when the LSM tree
+ * that is currently being built uses the new space format
+ * while other LSM trees still use the old space format.
+ */
+ if (*region_stmt == NULL || (*region_stmt)->format_id != format_id) {
*region_stmt = vy_stmt_dup_lsregion(stmt, &mem->env->allocator,
mem->generation);
if (*region_stmt == NULL)
@@ -788,7 +797,6 @@ vy_lsm_set(struct vy_lsm *lsm, struct vy_mem *mem,
lsm->stat.memory.count.bytes += tuple_size(stmt);
/* Abort transaction if format was changed by DDL */
- uint32_t format_id = stmt->format_id;
if (format_id != tuple_format_id(mem->format_with_colmask) &&
format_id != tuple_format_id(mem->format)) {
diag_set(ClientError, ER_TRANSACTION_CONFLICT);
diff --git a/src/box/vy_quota.h b/src/box/vy_quota.h
index 61a69008..d741c34a 100644
--- a/src/box/vy_quota.h
+++ b/src/box/vy_quota.h
@@ -177,6 +177,16 @@ vy_quota_use(struct vy_quota *q, size_t size, double timeout)
return 0;
}
+/**
+ * Block the caller until the quota is not exceeded.
+ */
+static inline void
+vy_quota_wait(struct vy_quota *q)
+{
+ while (q->used > q->limit)
+ fiber_cond_wait(&q->cond);
+}
+
#if defined(__cplusplus)
} /* extern "C" */
#endif /* defined(__cplusplus) */
diff --git a/src/box/vy_scheduler.c b/src/box/vy_scheduler.c
index e3f2e223..a82fe9f2 100644
--- a/src/box/vy_scheduler.c
+++ b/src/box/vy_scheduler.c
@@ -448,6 +448,35 @@ vy_scheduler_trigger_dump(struct vy_scheduler *scheduler)
fiber_cond_signal(&scheduler->scheduler_cond);
}
+int
+vy_scheduler_dump(struct vy_scheduler *scheduler)
+{
+ /*
+ * We must not start dump if checkpoint is in progress
+ * so first wait for checkpoint to complete.
+ */
+ while (scheduler->checkpoint_in_progress)
+ fiber_cond_wait(&scheduler->dump_cond);
+
+ /* Trigger dump. */
+ if (scheduler->generation == scheduler->dump_generation)
+ scheduler->dump_start = ev_monotonic_now(loop());
+ int64_t generation = ++scheduler->generation;
+ fiber_cond_signal(&scheduler->scheduler_cond);
+
+ /* Wait for dump to complete. */
+ while (scheduler->dump_generation < generation) {
+ if (scheduler->is_throttled) {
+ /* Dump error occurred. */
+ struct error *e = diag_last_error(&scheduler->diag);
+ diag_add_error(diag_get(), e);
+ return -1;
+ }
+ fiber_cond_wait(&scheduler->dump_cond);
+ }
+ return 0;
+}
+
void
vy_scheduler_force_compaction(struct vy_scheduler *scheduler,
struct vy_lsm *lsm)
diff --git a/src/box/vy_scheduler.h b/src/box/vy_scheduler.h
index d3bc4c03..777756c0 100644
--- a/src/box/vy_scheduler.h
+++ b/src/box/vy_scheduler.h
@@ -202,6 +202,13 @@ void
vy_scheduler_trigger_dump(struct vy_scheduler *scheduler);
/**
+ * Trigger dump of all currently existing in-memory trees
+ * and wait until it is complete. Returns 0 on success.
+ */
+int
+vy_scheduler_dump(struct vy_scheduler *scheduler);
+
+/**
* Force major compaction of an LSM tree.
*/
void
diff --git a/test/box/alter.result b/test/box/alter.result
index b3a14db6..eb7014d8 100644
--- a/test/box/alter.result
+++ b/test/box/alter.result
@@ -820,507 +820,6 @@ ts:drop()
---
...
--
--- gh-1557: NULL in indexes.
---
-NULL = require('msgpack').NULL
----
-...
-format = {}
----
-...
-format[1] = { name = 'field1', type = 'unsigned', is_nullable = true }
----
-...
-format[2] = { name = 'field2', type = 'unsigned', is_nullable = true }
----
-...
-s = box.schema.space.create('test', { format = format })
----
-...
-s:create_index('primary', { parts = { 'field1' } })
----
-- error: Primary index of the space 'test' can not contain nullable parts
-...
-s:create_index('primary', { parts = {{'field1', is_nullable = false}} })
----
-- error: Field 1 is nullable in space format, but not nullable in index parts
-...
-format[1].is_nullable = false
----
-...
-s:format(format)
----
-...
-s:create_index('primary', { parts = {{'field1', is_nullable = true}} })
----
-- error: Primary index of the space 'test' can not contain nullable parts
-...
-s:create_index('primary', { parts = {'field1'} })
----
-- unique: true
- parts:
- - type: unsigned
- is_nullable: false
- fieldno: 1
- id: 0
- space_id: 733
- name: primary
- type: TREE
-...
--- Check that is_nullable can't be set to false on non-empty space
-s:insert({1, NULL})
----
-- [1, null]
-...
-format[1].is_nullable = true
----
-...
-s:format(format)
----
-- error: Field 1 is nullable in space format, but not nullable in index parts
-...
-format[1].is_nullable = false
----
-...
-format[2].is_nullable = false
----
-...
-s:format(format)
----
-- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
-...
-s:delete(1)
----
-- [1, null]
-...
--- Disable is_nullable on empty space
-s:format(format)
----
-...
--- Disable is_nullable on a non-empty space.
-format[2].is_nullable = true
----
-...
-s:format(format)
----
-...
-s:replace{1, 1}
----
-- [1, 1]
-...
-format[2].is_nullable = false
----
-...
-s:format(format)
----
-...
--- Enable is_nullable on a non-empty space.
-format[2].is_nullable = true
----
-...
-s:format(format)
----
-...
-s:replace{1, box.NULL}
----
-- [1, null]
-...
-s:delete{1}
----
-- [1, null]
-...
-s:format({})
----
-...
-s:create_index('secondary', { parts = {{2, 'string', is_nullable = true}} })
----
-- unique: true
- parts:
- - type: string
- is_nullable: true
- fieldno: 2
- id: 1
- space_id: 733
- name: secondary
- type: TREE
-...
-s:insert({1, NULL})
----
-- [1, null]
-...
-s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
----
-- error: 'Tuple field 2 type does not match one required by operation: expected string'
-...
-s:delete({1})
----
-- [1, null]
-...
-s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
----
-...
-s:insert({1, NULL})
----
-- error: 'Tuple field 2 type does not match one required by operation: expected string'
-...
-s:insert({2, 'xxx'})
----
-- [2, 'xxx']
-...
-s.index.secondary:alter({ parts = {{2, 'string', is_nullable = true} }})
----
-...
-s:insert({1, NULL})
----
-- [1, null]
-...
-s:drop()
----
-...
-s = box.schema.create_space('test')
----
-...
-test_run:cmd("setopt delimiter ';'")
----
-- true
-...
-s:format({
- [1] = { name = 'id1', type = 'unsigned'},
- [2] = { name = 'id2', type = 'unsigned'},
- [3] = { name = 'id3', type = 'string'},
- [4] = { name = 'id4', type = 'string'},
- [5] = { name = 'id5', type = 'string'},
- [6] = { name = 'id6', type = 'string'},
-});
----
-...
-test_run:cmd("setopt delimiter ''");
----
-- true
-...
-s:format()
----
-- [{'name': 'id1', 'type': 'unsigned'}, {'name': 'id2', 'type': 'unsigned'}, {'name': 'id3',
- 'type': 'string'}, {'name': 'id4', 'type': 'string'}, {'name': 'id5', 'type': 'string'},
- {'name': 'id6', 'type': 'string'}]
-...
-_ = s:create_index('primary')
----
-...
-s:insert({1, 1, 'a', 'b', 'c', 'd'})
----
-- [1, 1, 'a', 'b', 'c', 'd']
-...
-s:drop()
----
-...
-s = box.schema.create_space('test')
----
-...
-idx = s:create_index('idx')
----
-...
-box.space.test == s
----
-- true
-...
-s:drop()
----
-...
---
--- gh-3000: index modifying must change key_def parts and
--- comparators. They can be changed, if there was compatible index
--- parts change. For example, a part type was changed from
--- unsigned to number. In such a case comparators must be reset
--- and part types updated.
---
-s = box.schema.create_space('test')
----
-...
-pk = s:create_index('pk')
----
-...
-s:replace{1}
----
-- [1]
-...
-pk:alter{parts = {{1, 'integer'}}}
----
-...
-s:replace{-2}
----
-- [-2]
-...
-s:select{}
----
-- - [-2]
- - [1]
-...
-s:drop()
----
-...
---
--- Allow to change is_nullable in index definition on non-empty
--- space.
---
-s = box.schema.create_space('test')
----
-...
-pk = s:create_index('pk')
----
-...
-sk1 = s:create_index('sk1', {parts = {{2, 'unsigned', is_nullable = true}}})
----
-...
-sk2 = s:create_index('sk2', {parts = {{3, 'unsigned', is_nullable = false}}})
----
-...
-s:replace{1, box.NULL, 1}
----
-- [1, null, 1]
-...
-sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
----
-- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
-...
-s:replace{1, 1, 1}
----
-- [1, 1, 1]
-...
-sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
----
-...
-s:replace{1, 1, box.NULL}
----
-- error: 'Tuple field 3 type does not match one required by operation: expected unsigned'
-...
-sk2:alter({parts = {{3, 'unsigned', is_nullable = true}}})
----
-...
-s:replace{1, 1, box.NULL}
----
-- [1, 1, null]
-...
-s:replace{2, 10, 100}
----
-- [2, 10, 100]
-...
-s:replace{3, 0, 20}
----
-- [3, 0, 20]
-...
-s:replace{4, 15, 150}
----
-- [4, 15, 150]
-...
-s:replace{5, 9, box.NULL}
----
-- [5, 9, null]
-...
-sk1:select{}
----
-- - [3, 0, 20]
- - [1, 1, null]
- - [5, 9, null]
- - [2, 10, 100]
- - [4, 15, 150]
-...
-sk2:select{}
----
-- - [1, 1, null]
- - [5, 9, null]
- - [3, 0, 20]
- - [2, 10, 100]
- - [4, 15, 150]
-...
-s:drop()
----
-...
---
--- gh-3008: allow multiple types on the same field.
---
-format = {}
----
-...
-format[1] = {name = 'field1', type = 'unsigned'}
----
-...
-format[2] = {name = 'field2', type = 'scalar'}
----
-...
-format[3] = {name = 'field3', type = 'integer'}
----
-...
-s = box.schema.create_space('test', {format = format})
----
-...
-pk = s:create_index('pk')
----
-...
-sk1 = s:create_index('sk1', {parts = {{2, 'number'}}})
----
-...
-sk2 = s:create_index('sk2', {parts = {{2, 'integer'}}})
----
-...
-sk3 = s:create_index('sk3', {parts = {{2, 'unsigned'}}})
----
-...
-sk4 = s:create_index('sk4', {parts = {{3, 'number'}}})
----
-...
-s:format()
----
-- [{'name': 'field1', 'type': 'unsigned'}, {'name': 'field2', 'type': 'scalar'}, {
- 'name': 'field3', 'type': 'integer'}]
-...
-s:replace{1, '100', -20.2}
----
-- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
-...
-s:replace{1, 100, -20.2}
----
-- error: 'Tuple field 3 type does not match one required by operation: expected integer'
-...
-s:replace{1, 100, -20}
----
-- [1, 100, -20]
-...
-s:replace{2, 50, 0}
----
-- [2, 50, 0]
-...
-s:replace{3, 150, -60}
----
-- [3, 150, -60]
-...
-s:replace{4, 0, 120}
----
-- [4, 0, 120]
-...
-pk:select{}
----
-- - [1, 100, -20]
- - [2, 50, 0]
- - [3, 150, -60]
- - [4, 0, 120]
-...
-sk1:select{}
----
-- - [4, 0, 120]
- - [2, 50, 0]
- - [1, 100, -20]
- - [3, 150, -60]
-...
-sk2:select{}
----
-- - [4, 0, 120]
- - [2, 50, 0]
- - [1, 100, -20]
- - [3, 150, -60]
-...
-sk3:select{}
----
-- - [4, 0, 120]
- - [2, 50, 0]
- - [1, 100, -20]
- - [3, 150, -60]
-...
-sk4:select{}
----
-- - [3, 150, -60]
- - [1, 100, -20]
- - [2, 50, 0]
- - [4, 0, 120]
-...
-sk1:alter{parts = {{2, 'unsigned'}}}
----
-...
-sk2:alter{parts = {{2, 'unsigned'}}}
----
-...
-sk4:alter{parts = {{3, 'integer'}}}
----
-...
-s:replace{1, 50.5, 1.5}
----
-- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
-...
-s:replace{1, 50, 1.5}
----
-- error: 'Tuple field 3 type does not match one required by operation: expected integer'
-...
-s:replace{5, 5, 5}
----
-- [5, 5, 5]
-...
-sk1:select{}
----
-- - [4, 0, 120]
- - [5, 5, 5]
- - [2, 50, 0]
- - [1, 100, -20]
- - [3, 150, -60]
-...
-sk2:select{}
----
-- - [4, 0, 120]
- - [5, 5, 5]
- - [2, 50, 0]
- - [1, 100, -20]
- - [3, 150, -60]
-...
-sk3:select{}
----
-- - [4, 0, 120]
- - [5, 5, 5]
- - [2, 50, 0]
- - [1, 100, -20]
- - [3, 150, -60]
-...
-sk4:select{}
----
-- - [3, 150, -60]
- - [1, 100, -20]
- - [2, 50, 0]
- - [5, 5, 5]
- - [4, 0, 120]
-...
-sk1:drop()
----
-...
-sk2:drop()
----
-...
-sk3:drop()
----
-...
--- Remove 'unsigned' constraints from indexes, and 'scalar' now
--- can be inserted in the second field.
-s:replace{1, true, 100}
----
-- [1, true, 100]
-...
-s:select{}
----
-- - [1, true, 100]
- - [2, 50, 0]
- - [3, 150, -60]
- - [4, 0, 120]
- - [5, 5, 5]
-...
-sk4:select{}
----
-- - [3, 150, -60]
- - [2, 50, 0]
- - [5, 5, 5]
- - [1, true, 100]
- - [4, 0, 120]
-...
-s:drop()
----
-...
---
-- gh-2914: Allow any space name which consists of printable characters
--
identifier = require("identifier")
diff --git a/test/box/alter.test.lua b/test/box/alter.test.lua
index 11b8125f..b2b25390 100644
--- a/test/box/alter.test.lua
+++ b/test/box/alter.test.lua
@@ -316,165 +316,6 @@ n
ts:drop()
--
--- gh-1557: NULL in indexes.
---
-
-NULL = require('msgpack').NULL
-
-format = {}
-format[1] = { name = 'field1', type = 'unsigned', is_nullable = true }
-format[2] = { name = 'field2', type = 'unsigned', is_nullable = true }
-s = box.schema.space.create('test', { format = format })
-s:create_index('primary', { parts = { 'field1' } })
-s:create_index('primary', { parts = {{'field1', is_nullable = false}} })
-format[1].is_nullable = false
-s:format(format)
-s:create_index('primary', { parts = {{'field1', is_nullable = true}} })
-
-s:create_index('primary', { parts = {'field1'} })
-
--- Check that is_nullable can't be set to false on non-empty space
-s:insert({1, NULL})
-format[1].is_nullable = true
-s:format(format)
-format[1].is_nullable = false
-format[2].is_nullable = false
-s:format(format)
-s:delete(1)
--- Disable is_nullable on empty space
-s:format(format)
--- Disable is_nullable on a non-empty space.
-format[2].is_nullable = true
-s:format(format)
-s:replace{1, 1}
-format[2].is_nullable = false
-s:format(format)
--- Enable is_nullable on a non-empty space.
-format[2].is_nullable = true
-s:format(format)
-s:replace{1, box.NULL}
-s:delete{1}
-s:format({})
-
-s:create_index('secondary', { parts = {{2, 'string', is_nullable = true}} })
-s:insert({1, NULL})
-s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
-s:delete({1})
-s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
-s:insert({1, NULL})
-s:insert({2, 'xxx'})
-s.index.secondary:alter({ parts = {{2, 'string', is_nullable = true} }})
-s:insert({1, NULL})
-
-s:drop()
-
-s = box.schema.create_space('test')
-test_run:cmd("setopt delimiter ';'")
-s:format({
- [1] = { name = 'id1', type = 'unsigned'},
- [2] = { name = 'id2', type = 'unsigned'},
- [3] = { name = 'id3', type = 'string'},
- [4] = { name = 'id4', type = 'string'},
- [5] = { name = 'id5', type = 'string'},
- [6] = { name = 'id6', type = 'string'},
-});
-test_run:cmd("setopt delimiter ''");
-s:format()
-_ = s:create_index('primary')
-s:insert({1, 1, 'a', 'b', 'c', 'd'})
-s:drop()
-
-s = box.schema.create_space('test')
-idx = s:create_index('idx')
-box.space.test == s
-s:drop()
-
---
--- gh-3000: index modifying must change key_def parts and
--- comparators. They can be changed, if there was compatible index
--- parts change. For example, a part type was changed from
--- unsigned to number. In such a case comparators must be reset
--- and part types updated.
---
-s = box.schema.create_space('test')
-pk = s:create_index('pk')
-s:replace{1}
-pk:alter{parts = {{1, 'integer'}}}
-s:replace{-2}
-s:select{}
-s:drop()
-
---
--- Allow to change is_nullable in index definition on non-empty
--- space.
---
-s = box.schema.create_space('test')
-pk = s:create_index('pk')
-sk1 = s:create_index('sk1', {parts = {{2, 'unsigned', is_nullable = true}}})
-sk2 = s:create_index('sk2', {parts = {{3, 'unsigned', is_nullable = false}}})
-s:replace{1, box.NULL, 1}
-sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
-s:replace{1, 1, 1}
-sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
-s:replace{1, 1, box.NULL}
-sk2:alter({parts = {{3, 'unsigned', is_nullable = true}}})
-s:replace{1, 1, box.NULL}
-s:replace{2, 10, 100}
-s:replace{3, 0, 20}
-s:replace{4, 15, 150}
-s:replace{5, 9, box.NULL}
-sk1:select{}
-sk2:select{}
-s:drop()
-
---
--- gh-3008: allow multiple types on the same field.
---
-format = {}
-format[1] = {name = 'field1', type = 'unsigned'}
-format[2] = {name = 'field2', type = 'scalar'}
-format[3] = {name = 'field3', type = 'integer'}
-s = box.schema.create_space('test', {format = format})
-pk = s:create_index('pk')
-sk1 = s:create_index('sk1', {parts = {{2, 'number'}}})
-sk2 = s:create_index('sk2', {parts = {{2, 'integer'}}})
-sk3 = s:create_index('sk3', {parts = {{2, 'unsigned'}}})
-sk4 = s:create_index('sk4', {parts = {{3, 'number'}}})
-s:format()
-s:replace{1, '100', -20.2}
-s:replace{1, 100, -20.2}
-s:replace{1, 100, -20}
-s:replace{2, 50, 0}
-s:replace{3, 150, -60}
-s:replace{4, 0, 120}
-pk:select{}
-sk1:select{}
-sk2:select{}
-sk3:select{}
-sk4:select{}
-
-sk1:alter{parts = {{2, 'unsigned'}}}
-sk2:alter{parts = {{2, 'unsigned'}}}
-sk4:alter{parts = {{3, 'integer'}}}
-s:replace{1, 50.5, 1.5}
-s:replace{1, 50, 1.5}
-s:replace{5, 5, 5}
-sk1:select{}
-sk2:select{}
-sk3:select{}
-sk4:select{}
-
-sk1:drop()
-sk2:drop()
-sk3:drop()
--- Remove 'unsigned' constraints from indexes, and 'scalar' now
--- can be inserted in the second field.
-s:replace{1, true, 100}
-s:select{}
-sk4:select{}
-s:drop()
-
---
-- gh-2914: Allow any space name which consists of printable characters
--
identifier = require("identifier")
diff --git a/test/engine/ddl.result b/test/engine/ddl.result
index 04062ac1..6c915879 100644
--- a/test/engine/ddl.result
+++ b/test/engine/ddl.result
@@ -1352,3 +1352,675 @@ s:select()
s:drop()
---
...
+--
+-- gh-1557: NULL in indexes.
+--
+NULL = require('msgpack').NULL
+---
+...
+format = {}
+---
+...
+format[1] = { name = 'field1', type = 'unsigned', is_nullable = true }
+---
+...
+format[2] = { name = 'field2', type = 'unsigned', is_nullable = true }
+---
+...
+s = box.schema.space.create('test', {engine = engine, format = format})
+---
+...
+s:create_index('primary', { parts = { 'field1' } })
+---
+- error: Primary index of the space 'test' can not contain nullable parts
+...
+s:create_index('primary', { parts = {{'field1', is_nullable = false}} })
+---
+- error: Field 1 is nullable in space format, but not nullable in index parts
+...
+format[1].is_nullable = false
+---
+...
+s:format(format)
+---
+...
+s:create_index('primary', { parts = {{'field1', is_nullable = true}} })
+---
+- error: Primary index of the space 'test' can not contain nullable parts
+...
+i = s:create_index('primary', { parts = {'field1'} })
+---
+...
+i.parts
+---
+- - type: unsigned
+ is_nullable: false
+ fieldno: 1
+...
+-- Check that is_nullable can't be set to false on non-empty space
+s:insert({1, NULL})
+---
+- [1, null]
+...
+format[1].is_nullable = true
+---
+...
+s:format(format)
+---
+- error: Field 1 is nullable in space format, but not nullable in index parts
+...
+format[1].is_nullable = false
+---
+...
+format[2].is_nullable = false
+---
+...
+s:format(format)
+---
+- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
+...
+_ = s:delete(1)
+---
+...
+-- Disable is_nullable on empty space
+s:format(format)
+---
+...
+-- Disable is_nullable on a non-empty space.
+format[2].is_nullable = true
+---
+...
+s:format(format)
+---
+...
+s:replace{1, 1}
+---
+- [1, 1]
+...
+format[2].is_nullable = false
+---
+...
+s:format(format)
+---
+...
+-- Enable is_nullable on a non-empty space.
+format[2].is_nullable = true
+---
+...
+s:format(format)
+---
+...
+s:replace{1, box.NULL}
+---
+- [1, null]
+...
+_ = s:delete{1}
+---
+...
+s:format({})
+---
+...
+i = s:create_index('secondary', { parts = {{2, 'string', is_nullable = true}} })
+---
+...
+i.parts
+---
+- - type: string
+ is_nullable: true
+ fieldno: 2
+...
+s:insert({1, NULL})
+---
+- [1, null]
+...
+s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
+---
+- error: 'Tuple field 2 type does not match one required by operation: expected string'
+...
+_ = s:delete({1})
+---
+...
+s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
+---
+...
+s:insert({1, NULL})
+---
+- error: 'Tuple field 2 type does not match one required by operation: expected string'
+...
+s:insert({2, 'xxx'})
+---
+- [2, 'xxx']
+...
+s.index.secondary:alter({ parts = {{2, 'string', is_nullable = true} }})
+---
+...
+s:insert({1, NULL})
+---
+- [1, null]
+...
+s:drop()
+---
+...
+s = box.schema.space.create('test', {engine = engine})
+---
+...
+inspector:cmd("setopt delimiter ';'")
+---
+- true
+...
+s:format({
+ [1] = { name = 'id1', type = 'unsigned'},
+ [2] = { name = 'id2', type = 'unsigned'},
+ [3] = { name = 'id3', type = 'string'},
+ [4] = { name = 'id4', type = 'string'},
+ [5] = { name = 'id5', type = 'string'},
+ [6] = { name = 'id6', type = 'string'},
+});
+---
+...
+inspector:cmd("setopt delimiter ''");
+---
+- true
+...
+s:format()
+---
+- [{'name': 'id1', 'type': 'unsigned'}, {'name': 'id2', 'type': 'unsigned'}, {'name': 'id3',
+ 'type': 'string'}, {'name': 'id4', 'type': 'string'}, {'name': 'id5', 'type': 'string'},
+ {'name': 'id6', 'type': 'string'}]
+...
+_ = s:create_index('primary')
+---
+...
+s:insert({1, 1, 'a', 'b', 'c', 'd'})
+---
+- [1, 1, 'a', 'b', 'c', 'd']
+...
+s:drop()
+---
+...
+s = box.schema.space.create('test', {engine = engine})
+---
+...
+idx = s:create_index('idx')
+---
+...
+box.space.test == s
+---
+- true
+...
+s:drop()
+---
+...
+--
+-- gh-3000: index modifying must change key_def parts and
+-- comparators. They can be changed, if there was compatible index
+-- parts change. For example, a part type was changed from
+-- unsigned to number. In such a case comparators must be reset
+-- and part types updated.
+--
+s = box.schema.space.create('test', {engine = engine})
+---
+...
+pk = s:create_index('pk')
+---
+...
+s:replace{1}
+---
+- [1]
+...
+pk:alter{parts = {{1, 'integer'}}}
+---
+...
+s:replace{-2}
+---
+- [-2]
+...
+s:select{}
+---
+- - [-2]
+ - [1]
+...
+s:drop()
+---
+...
+--
+-- Allow to change is_nullable in index definition on non-empty
+-- space.
+--
+s = box.schema.space.create('test', {engine = engine})
+---
+...
+pk = s:create_index('pk')
+---
+...
+sk1 = s:create_index('sk1', {parts = {{2, 'unsigned', is_nullable = true}}})
+---
+...
+sk2 = s:create_index('sk2', {parts = {{3, 'unsigned', is_nullable = false}}})
+---
+...
+s:replace{1, box.NULL, 1}
+---
+- [1, null, 1]
+...
+sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
+---
+- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
+...
+s:replace{1, 1, 1}
+---
+- [1, 1, 1]
+...
+sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
+---
+...
+s:replace{1, 1, box.NULL}
+---
+- error: 'Tuple field 3 type does not match one required by operation: expected unsigned'
+...
+sk2:alter({parts = {{3, 'unsigned', is_nullable = true}}})
+---
+...
+s:replace{1, 1, box.NULL}
+---
+- [1, 1, null]
+...
+s:replace{2, 10, 100}
+---
+- [2, 10, 100]
+...
+s:replace{3, 0, 20}
+---
+- [3, 0, 20]
+...
+s:replace{4, 15, 150}
+---
+- [4, 15, 150]
+...
+s:replace{5, 9, box.NULL}
+---
+- [5, 9, null]
+...
+sk1:select{}
+---
+- - [3, 0, 20]
+ - [1, 1, null]
+ - [5, 9, null]
+ - [2, 10, 100]
+ - [4, 15, 150]
+...
+sk2:select{}
+---
+- - [1, 1, null]
+ - [5, 9, null]
+ - [3, 0, 20]
+ - [2, 10, 100]
+ - [4, 15, 150]
+...
+s:drop()
+---
+...
+--
+-- gh-3008: allow multiple types on the same field.
+--
+format = {}
+---
+...
+format[1] = {name = 'field1', type = 'unsigned'}
+---
+...
+format[2] = {name = 'field2', type = 'scalar'}
+---
+...
+format[3] = {name = 'field3', type = 'integer'}
+---
+...
+s = box.schema.space.create('test', {engine = engine, format = format})
+---
+...
+pk = s:create_index('pk')
+---
+...
+sk1 = s:create_index('sk1', {parts = {{2, 'number'}}})
+---
+...
+sk2 = s:create_index('sk2', {parts = {{2, 'integer'}}})
+---
+...
+sk3 = s:create_index('sk3', {parts = {{2, 'unsigned'}}})
+---
+...
+sk4 = s:create_index('sk4', {parts = {{3, 'number'}}})
+---
+...
+s:format()
+---
+- [{'name': 'field1', 'type': 'unsigned'}, {'name': 'field2', 'type': 'scalar'}, {
+ 'name': 'field3', 'type': 'integer'}]
+...
+s:replace{1, '100', -20.2}
+---
+- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
+...
+s:replace{1, 100, -20.2}
+---
+- error: 'Tuple field 3 type does not match one required by operation: expected integer'
+...
+s:replace{1, 100, -20}
+---
+- [1, 100, -20]
+...
+s:replace{2, 50, 0}
+---
+- [2, 50, 0]
+...
+s:replace{3, 150, -60}
+---
+- [3, 150, -60]
+...
+s:replace{4, 0, 120}
+---
+- [4, 0, 120]
+...
+pk:select{}
+---
+- - [1, 100, -20]
+ - [2, 50, 0]
+ - [3, 150, -60]
+ - [4, 0, 120]
+...
+sk1:select{}
+---
+- - [4, 0, 120]
+ - [2, 50, 0]
+ - [1, 100, -20]
+ - [3, 150, -60]
+...
+sk2:select{}
+---
+- - [4, 0, 120]
+ - [2, 50, 0]
+ - [1, 100, -20]
+ - [3, 150, -60]
+...
+sk3:select{}
+---
+- - [4, 0, 120]
+ - [2, 50, 0]
+ - [1, 100, -20]
+ - [3, 150, -60]
+...
+sk4:select{}
+---
+- - [3, 150, -60]
+ - [1, 100, -20]
+ - [2, 50, 0]
+ - [4, 0, 120]
+...
+sk1:alter{parts = {{2, 'unsigned'}}}
+---
+...
+sk2:alter{parts = {{2, 'unsigned'}}}
+---
+...
+sk4:alter{parts = {{3, 'integer'}}}
+---
+...
+s:replace{1, 50.5, 1.5}
+---
+- error: 'Tuple field 2 type does not match one required by operation: expected unsigned'
+...
+s:replace{1, 50, 1.5}
+---
+- error: 'Tuple field 3 type does not match one required by operation: expected integer'
+...
+s:replace{5, 5, 5}
+---
+- [5, 5, 5]
+...
+sk1:select{}
+---
+- - [4, 0, 120]
+ - [5, 5, 5]
+ - [2, 50, 0]
+ - [1, 100, -20]
+ - [3, 150, -60]
+...
+sk2:select{}
+---
+- - [4, 0, 120]
+ - [5, 5, 5]
+ - [2, 50, 0]
+ - [1, 100, -20]
+ - [3, 150, -60]
+...
+sk3:select{}
+---
+- - [4, 0, 120]
+ - [5, 5, 5]
+ - [2, 50, 0]
+ - [1, 100, -20]
+ - [3, 150, -60]
+...
+sk4:select{}
+---
+- - [3, 150, -60]
+ - [1, 100, -20]
+ - [2, 50, 0]
+ - [5, 5, 5]
+ - [4, 0, 120]
+...
+sk1:drop()
+---
+...
+sk2:drop()
+---
+...
+sk3:drop()
+---
+...
+-- Remove 'unsigned' constraints from indexes, and 'scalar' now
+-- can be inserted in the second field.
+s:replace{1, true, 100}
+---
+- [1, true, 100]
+...
+s:select{}
+---
+- - [1, true, 100]
+ - [2, 50, 0]
+ - [3, 150, -60]
+ - [4, 0, 120]
+ - [5, 5, 5]
+...
+sk4:select{}
+---
+- - [3, 150, -60]
+ - [2, 50, 0]
+ - [5, 5, 5]
+ - [1, true, 100]
+ - [4, 0, 120]
+...
+s:drop()
+---
+...
+--
+-- Creating/altering a secondary index of a non-empty space.
+--
+s = box.schema.space.create('test', {engine = engine})
+---
+...
+_ = s:create_index('pk')
+---
+...
+_ = s:insert{1, 'zzz', 'aaa', 999}
+---
+...
+_ = s:insert{2, 'yyy', 'bbb', 888}
+---
+...
+_ = s:insert{3, 'xxx', 'ccc', 777}
+---
+...
+box.snapshot()
+---
+- ok
+...
+_ = s:update(1, {{'!', -1, 'eee'}})
+---
+...
+_ = s:upsert({2, '2', '2', -2}, {{'=', 4, -888}})
+---
+...
+_ = s:replace(s:get(3):update{{'=', 3, box.NULL}})
+---
+...
+_ = s:upsert({4, 'zzz', 'ddd', -666}, {{'!', -1, 'abc'}})
+---
+...
+box.snapshot()
+---
+- ok
+...
+_ = s:update(1, {{'=', 5, 'fff'}})
+---
+...
+_ = s:upsert({3, '3', '3', -3}, {{'=', 5, 'ggg'}})
+---
+...
+_ = s:insert{5, 'xxx', 'eee', 555, 'hhh'}
+---
+...
+_ = s:replace{6, 'yyy', box.NULL, -444}
+---
+...
+s:select()
+---
+- - [1, 'zzz', 'aaa', 999, 'fff']
+ - [2, 'yyy', 'bbb', -888]
+ - [3, 'xxx', null, 777, 'ggg']
+ - [4, 'zzz', 'ddd', -666]
+ - [5, 'xxx', 'eee', 555, 'hhh']
+ - [6, 'yyy', null, -444]
+...
+s:create_index('sk', {parts = {2, 'string'}}) -- error: unique constraint
+---
+- error: Duplicate key exists in unique index 'sk' in space 'test'
+...
+s:create_index('sk', {parts = {3, 'string'}}) -- error: nullability constraint
+---
+- error: 'Tuple field 3 type does not match one required by operation: expected string'
+...
+s:create_index('sk', {parts = {4, 'unsigned'}}) -- error: field type
+---
+- error: 'Tuple field 4 type does not match one required by operation: expected unsigned'
+...
+s:create_index('sk', {parts = {4, 'integer', 5, 'string'}}) -- error: field missing
+---
+- error: Tuple field count 4 is less than required by space format or defined indexes
+ (expected at least 5)
+...
+i1 = s:create_index('i1', {parts = {2, 'string'}, unique = false})
+---
+...
+i2 = s:create_index('i2', {parts = {{3, 'string', is_nullable = true}}})
+---
+...
+i3 = s:create_index('i3', {parts = {4, 'integer'}})
+---
+...
+i1:select()
+---
+- - [3, 'xxx', null, 777, 'ggg']
+ - [5, 'xxx', 'eee', 555, 'hhh']
+ - [2, 'yyy', 'bbb', -888]
+ - [6, 'yyy', null, -444]
+ - [1, 'zzz', 'aaa', 999, 'fff']
+ - [4, 'zzz', 'ddd', -666]
+...
+i2:select()
+---
+- - [3, 'xxx', null, 777, 'ggg']
+ - [6, 'yyy', null, -444]
+ - [1, 'zzz', 'aaa', 999, 'fff']
+ - [2, 'yyy', 'bbb', -888]
+ - [4, 'zzz', 'ddd', -666]
+ - [5, 'xxx', 'eee', 555, 'hhh']
+...
+i3:select()
+---
+- - [2, 'yyy', 'bbb', -888]
+ - [4, 'zzz', 'ddd', -666]
+ - [6, 'yyy', null, -444]
+ - [5, 'xxx', 'eee', 555, 'hhh']
+ - [3, 'xxx', null, 777, 'ggg']
+ - [1, 'zzz', 'aaa', 999, 'fff']
+...
+i1:alter{unique = true} -- error: unique contraint
+---
+- error: Duplicate key exists in unique index 'i1' in space 'test'
+...
+i2:alter{parts = {3, 'string'}} -- error: nullability contraint
+---
+- error: 'Tuple field 3 type does not match one required by operation: expected string'
+...
+i3:alter{parts = {4, 'unsigned'}} -- error: field type
+---
+- error: 'Tuple field 4 type does not match one required by operation: expected unsigned'
+...
+i3:alter{parts = {4, 'integer', 5, 'string'}} -- error: field missing
+---
+- error: Tuple field count 4 is less than required by space format or defined indexes
+ (expected at least 5)
+...
+i3:alter{parts = {2, 'string', 4, 'integer'}} -- ok
+---
+...
+i3:select()
+---
+- - [5, 'xxx', 'eee', 555, 'hhh']
+ - [3, 'xxx', null, 777, 'ggg']
+ - [2, 'yyy', 'bbb', -888]
+ - [6, 'yyy', null, -444]
+ - [4, 'zzz', 'ddd', -666]
+ - [1, 'zzz', 'aaa', 999, 'fff']
+...
+-- Check that recovery works.
+inspector:cmd("restart server default")
+s = box.space.test
+---
+...
+s.index.i1:select()
+---
+- - [3, 'xxx', null, 777, 'ggg']
+ - [5, 'xxx', 'eee', 555, 'hhh']
+ - [2, 'yyy', 'bbb', -888]
+ - [6, 'yyy', null, -444]
+ - [1, 'zzz', 'aaa', 999, 'fff']
+ - [4, 'zzz', 'ddd', -666]
+...
+s.index.i2:select()
+---
+- - [3, 'xxx', null, 777, 'ggg']
+ - [6, 'yyy', null, -444]
+ - [1, 'zzz', 'aaa', 999, 'fff']
+ - [2, 'yyy', 'bbb', -888]
+ - [4, 'zzz', 'ddd', -666]
+ - [5, 'xxx', 'eee', 555, 'hhh']
+...
+s.index.i3:select()
+---
+- - [5, 'xxx', 'eee', 555, 'hhh']
+ - [3, 'xxx', null, 777, 'ggg']
+ - [2, 'yyy', 'bbb', -888]
+ - [6, 'yyy', null, -444]
+ - [4, 'zzz', 'ddd', -666]
+ - [1, 'zzz', 'aaa', 999, 'fff']
+...
+box.snapshot()
+---
+- ok
+...
+s:drop()
+---
+...
diff --git a/test/engine/ddl.test.lua b/test/engine/ddl.test.lua
index f3d68e1b..2593535b 100644
--- a/test/engine/ddl.test.lua
+++ b/test/engine/ddl.test.lua
@@ -498,3 +498,224 @@ s:format(format)
s:format()
s:select()
s:drop()
+
+--
+-- gh-1557: NULL in indexes.
+--
+
+NULL = require('msgpack').NULL
+
+format = {}
+format[1] = { name = 'field1', type = 'unsigned', is_nullable = true }
+format[2] = { name = 'field2', type = 'unsigned', is_nullable = true }
+s = box.schema.space.create('test', {engine = engine, format = format})
+s:create_index('primary', { parts = { 'field1' } })
+s:create_index('primary', { parts = {{'field1', is_nullable = false}} })
+format[1].is_nullable = false
+s:format(format)
+s:create_index('primary', { parts = {{'field1', is_nullable = true}} })
+
+i = s:create_index('primary', { parts = {'field1'} })
+i.parts
+
+-- Check that is_nullable can't be set to false on non-empty space
+s:insert({1, NULL})
+format[1].is_nullable = true
+s:format(format)
+format[1].is_nullable = false
+format[2].is_nullable = false
+s:format(format)
+_ = s:delete(1)
+-- Disable is_nullable on empty space
+s:format(format)
+-- Disable is_nullable on a non-empty space.
+format[2].is_nullable = true
+s:format(format)
+s:replace{1, 1}
+format[2].is_nullable = false
+s:format(format)
+-- Enable is_nullable on a non-empty space.
+format[2].is_nullable = true
+s:format(format)
+s:replace{1, box.NULL}
+_ = s:delete{1}
+s:format({})
+
+i = s:create_index('secondary', { parts = {{2, 'string', is_nullable = true}} })
+i.parts
+
+s:insert({1, NULL})
+s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
+_ = s:delete({1})
+s.index.secondary:alter({ parts = {{2, 'string', is_nullable = false} }})
+s:insert({1, NULL})
+s:insert({2, 'xxx'})
+s.index.secondary:alter({ parts = {{2, 'string', is_nullable = true} }})
+s:insert({1, NULL})
+
+s:drop()
+
+s = box.schema.space.create('test', {engine = engine})
+inspector:cmd("setopt delimiter ';'")
+s:format({
+ [1] = { name = 'id1', type = 'unsigned'},
+ [2] = { name = 'id2', type = 'unsigned'},
+ [3] = { name = 'id3', type = 'string'},
+ [4] = { name = 'id4', type = 'string'},
+ [5] = { name = 'id5', type = 'string'},
+ [6] = { name = 'id6', type = 'string'},
+});
+inspector:cmd("setopt delimiter ''");
+s:format()
+_ = s:create_index('primary')
+s:insert({1, 1, 'a', 'b', 'c', 'd'})
+s:drop()
+
+s = box.schema.space.create('test', {engine = engine})
+idx = s:create_index('idx')
+box.space.test == s
+s:drop()
+
+--
+-- gh-3000: index modifying must change key_def parts and
+-- comparators. They can be changed, if there was compatible index
+-- parts change. For example, a part type was changed from
+-- unsigned to number. In such a case comparators must be reset
+-- and part types updated.
+--
+s = box.schema.space.create('test', {engine = engine})
+pk = s:create_index('pk')
+s:replace{1}
+pk:alter{parts = {{1, 'integer'}}}
+s:replace{-2}
+s:select{}
+s:drop()
+
+--
+-- Allow to change is_nullable in index definition on non-empty
+-- space.
+--
+s = box.schema.space.create('test', {engine = engine})
+pk = s:create_index('pk')
+sk1 = s:create_index('sk1', {parts = {{2, 'unsigned', is_nullable = true}}})
+sk2 = s:create_index('sk2', {parts = {{3, 'unsigned', is_nullable = false}}})
+s:replace{1, box.NULL, 1}
+sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
+s:replace{1, 1, 1}
+sk1:alter({parts = {{2, 'unsigned', is_nullable = false}}})
+s:replace{1, 1, box.NULL}
+sk2:alter({parts = {{3, 'unsigned', is_nullable = true}}})
+s:replace{1, 1, box.NULL}
+s:replace{2, 10, 100}
+s:replace{3, 0, 20}
+s:replace{4, 15, 150}
+s:replace{5, 9, box.NULL}
+sk1:select{}
+sk2:select{}
+s:drop()
+
+--
+-- gh-3008: allow multiple types on the same field.
+--
+format = {}
+format[1] = {name = 'field1', type = 'unsigned'}
+format[2] = {name = 'field2', type = 'scalar'}
+format[3] = {name = 'field3', type = 'integer'}
+s = box.schema.space.create('test', {engine = engine, format = format})
+pk = s:create_index('pk')
+sk1 = s:create_index('sk1', {parts = {{2, 'number'}}})
+sk2 = s:create_index('sk2', {parts = {{2, 'integer'}}})
+sk3 = s:create_index('sk3', {parts = {{2, 'unsigned'}}})
+sk4 = s:create_index('sk4', {parts = {{3, 'number'}}})
+s:format()
+s:replace{1, '100', -20.2}
+s:replace{1, 100, -20.2}
+s:replace{1, 100, -20}
+s:replace{2, 50, 0}
+s:replace{3, 150, -60}
+s:replace{4, 0, 120}
+pk:select{}
+sk1:select{}
+sk2:select{}
+sk3:select{}
+sk4:select{}
+
+sk1:alter{parts = {{2, 'unsigned'}}}
+sk2:alter{parts = {{2, 'unsigned'}}}
+sk4:alter{parts = {{3, 'integer'}}}
+s:replace{1, 50.5, 1.5}
+s:replace{1, 50, 1.5}
+s:replace{5, 5, 5}
+sk1:select{}
+sk2:select{}
+sk3:select{}
+sk4:select{}
+
+sk1:drop()
+sk2:drop()
+sk3:drop()
+-- Remove 'unsigned' constraints from indexes, and 'scalar' now
+-- can be inserted in the second field.
+s:replace{1, true, 100}
+s:select{}
+sk4:select{}
+s:drop()
+
+--
+-- Creating/altering a secondary index of a non-empty space.
+--
+s = box.schema.space.create('test', {engine = engine})
+_ = s:create_index('pk')
+
+_ = s:insert{1, 'zzz', 'aaa', 999}
+_ = s:insert{2, 'yyy', 'bbb', 888}
+_ = s:insert{3, 'xxx', 'ccc', 777}
+
+box.snapshot()
+
+_ = s:update(1, {{'!', -1, 'eee'}})
+_ = s:upsert({2, '2', '2', -2}, {{'=', 4, -888}})
+_ = s:replace(s:get(3):update{{'=', 3, box.NULL}})
+_ = s:upsert({4, 'zzz', 'ddd', -666}, {{'!', -1, 'abc'}})
+
+box.snapshot()
+
+_ = s:update(1, {{'=', 5, 'fff'}})
+_ = s:upsert({3, '3', '3', -3}, {{'=', 5, 'ggg'}})
+_ = s:insert{5, 'xxx', 'eee', 555, 'hhh'}
+_ = s:replace{6, 'yyy', box.NULL, -444}
+
+s:select()
+
+s:create_index('sk', {parts = {2, 'string'}}) -- error: unique constraint
+s:create_index('sk', {parts = {3, 'string'}}) -- error: nullability constraint
+s:create_index('sk', {parts = {4, 'unsigned'}}) -- error: field type
+s:create_index('sk', {parts = {4, 'integer', 5, 'string'}}) -- error: field missing
+
+i1 = s:create_index('i1', {parts = {2, 'string'}, unique = false})
+i2 = s:create_index('i2', {parts = {{3, 'string', is_nullable = true}}})
+i3 = s:create_index('i3', {parts = {4, 'integer'}})
+
+i1:select()
+i2:select()
+i3:select()
+
+i1:alter{unique = true} -- error: unique contraint
+i2:alter{parts = {3, 'string'}} -- error: nullability contraint
+i3:alter{parts = {4, 'unsigned'}} -- error: field type
+i3:alter{parts = {4, 'integer', 5, 'string'}} -- error: field missing
+
+i3:alter{parts = {2, 'string', 4, 'integer'}} -- ok
+i3:select()
+
+-- Check that recovery works.
+inspector:cmd("restart server default")
+
+s = box.space.test
+s.index.i1:select()
+s.index.i2:select()
+s.index.i3:select()
+
+box.snapshot()
+
+s:drop()
diff --git a/test/vinyl/ddl.result b/test/vinyl/ddl.result
index 3fdeea0e..16ee7097 100644
--- a/test/vinyl/ddl.result
+++ b/test/vinyl/ddl.result
@@ -68,202 +68,61 @@ index = space:create_index('primary', {type = 'hash'})
space:drop()
---
...
--- creation of a new index and altering the definition of an existing
--- index are unsupported for non-empty spaces
+-- rebuild of the primary index is not supported
space = box.schema.space.create('test', { engine = 'vinyl' })
---
...
-index = space:create_index('primary')
+pk = space:create_index('pk', {run_count_per_level = 1, run_size_ratio = 10})
---
...
-space:insert({1})
+space:replace{1, 1}
---
-- [1]
-...
--- fail because of wrong tuple format {1}, but need {1, ...}
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
----
-- error: Vinyl does not support building an index for a non-empty space
-...
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
----
-- error: Vinyl does not support building an index for a non-empty space
-...
-#box.space._index:select({space.id})
----
-- 1
-...
-box.space._index:get{space.id, 0}[6]
----
-- [[0, 'unsigned']]
-...
-space:drop()
----
-...
-space = box.schema.space.create('test', { engine = 'vinyl' })
----
-...
-index = space:create_index('primary')
----
-...
-space:insert({1, 2})
----
-- [1, 2]
-...
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
----
-- error: Vinyl does not support building an index for a non-empty space
-...
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
----
-- error: Vinyl does not support building an index for a non-empty space
-...
-#box.space._index:select({space.id})
----
-- 1
-...
-box.space._index:get{space.id, 0}[6]
----
-- [[0, 'unsigned']]
-...
-space:drop()
----
-...
-space = box.schema.space.create('test', { engine = 'vinyl' })
----
-...
-index = space:create_index('primary')
----
-...
-space:insert({1, 2})
----
-- [1, 2]
-...
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
----
-- error: Vinyl does not support building an index for a non-empty space
-...
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
----
-- error: Vinyl does not support building an index for a non-empty space
-...
-#box.space._index:select({space.id})
----
-- 1
-...
-box.space._index:get{space.id, 0}[6]
----
-- [[0, 'unsigned']]
-...
-space:delete({1})
----
-...
--- must fail because vy_mems have data
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
----
-- error: Vinyl does not support building an index for a non-empty space
+- [1, 1]
...
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
+pk:alter{parts = {2, 'unsigned'}} -- error: mem not empty
---
-- error: Vinyl does not support building an index for a non-empty space
+- error: Vinyl does not support rebuilding the primary index of a non-empty space
...
box.snapshot()
---
- ok
...
-while space.index.primary:stat().rows ~= 0 do fiber.sleep(0.01) end
+pk:alter{parts = {2, 'unsigned'}} -- error: run not empty
---
+- error: Vinyl does not support rebuilding the primary index of a non-empty space
...
--- after a dump REPLACE + DELETE = nothing, so the space is empty now and
--- can be altered.
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
----
-...
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
----
-...
-#box.space._index:select({space.id})
----
-- 2
-...
-box.space._index:get{space.id, 0}[6]
----
-- [[0, 'unsigned'], [1, 'unsigned']]
-...
-space:insert({1, 2})
----
-- [1, 2]
-...
-index:select{}
----
-- - [1, 2]
-...
-index2:select{}
----
-- - [1, 2]
-...
-space:drop()
----
-...
-space = box.schema.space.create('test', { engine = 'vinyl' })
+space:replace{2, 2}
---
+- [2, 2]
...
-index = space:create_index('primary', { run_count_per_level = 2 })
+space:delete{1}
---
...
-space:insert({1, 2})
+space:delete{2}
---
-- [1, 2]
-...
-box.snapshot()
----
-- ok
...
-space:delete({1})
+pk:alter{parts = {2, 'unsigned'}} -- error: mem/run not empty
---
+- error: Vinyl does not support rebuilding the primary index of a non-empty space
...
box.snapshot()
---
- ok
...
-while space.index.primary:stat().run_count ~= 2 do fiber.sleep(0.01) end
+-- wait for compaction to complete
+while pk:stat().disk.compact.count == 0 do fiber.sleep(0.01) end
---
...
--- must fail because vy_runs have data
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
+pk:alter{parts = {2, 'unsigned'}} -- success: space is empty now
---
-- error: Vinyl does not support building an index for a non-empty space
...
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
----
-- error: Vinyl does not support building an index for a non-empty space
-...
--- After compaction the REPLACE + DELETE + DELETE = nothing, so
--- the space is now empty and can be altered.
-space:delete({1})
----
-...
--- Make sure the run is big enough to trigger compaction.
-space:replace({2, 3})
----
-- [2, 3]
-...
-space:delete({2})
----
-...
-box.snapshot()
----
-- ok
-...
--- Wait until the dump is finished.
-while space.index.primary:stat().rows ~= 0 do fiber.sleep(0.01) end
----
-...
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
+space:replace{1, 2}
---
+- [1, 2]
...
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
+space:get(2)
---
+- [1, 2]
...
space:drop()
---
@@ -291,7 +150,7 @@ space:auto_increment{3}
...
box.space._index:replace{space.id, 0, 'pk', 'tree', {unique=true}, {{0, 'unsigned'}, {1, 'unsigned'}}}
---
-- error: Vinyl does not support building an index for a non-empty space
+- error: Vinyl does not support rebuilding the primary index of a non-empty space
...
space:select{}
---
@@ -726,72 +585,264 @@ box.space.test.index.pk
box.space.test:drop()
---
...
--- gh-2449 change 'unique' index property from true to false
-s = box.schema.space.create('test', { engine = 'vinyl' })
+-- Narrowing indexed field type entails index rebuild.
+s = box.schema.space.create('test', {engine = 'vinyl'})
+---
+...
+pk = s:create_index('pk', {parts = {1, 'unsigned'}})
---
...
-_ = s:create_index('primary')
+s:replace{1}
---
+- [1]
...
-_ = s:create_index('secondary', {unique = true, parts = {2, 'unsigned'}})
+-- Extending field type is allowed without rebuild.
+pk:alter{parts = {1, 'integer'}}
---
...
-s:insert{1, 10}
+-- Should fail as we do not support rebuilding the primary index of a non-empty space.
+pk:alter{parts = {1, 'unsigned'}}
---
-- [1, 10]
+- error: Vinyl does not support rebuilding the primary index of a non-empty space
...
-s.index.secondary:alter{unique = false} -- ok
+s:replace{-1}
---
+- [-1]
...
-s.index.secondary.unique
+s:drop()
---
-- false
...
-s.index.secondary:alter{unique = true} -- error
+--
+-- Check that all modifications done to the space during index build
+-- are reflected in the new index.
+--
+math.randomseed(os.time())
---
-- error: Vinyl does not support building an index for a non-empty space
...
-s.index.secondary.unique
+s = box.schema.space.create('test', {engine = 'vinyl'})
---
-- false
...
-s:insert{2, 10}
+_ = s:create_index('pk')
---
-- [2, 10]
...
-s.index.secondary:select(10)
+test_run:cmd("setopt delimiter ';'")
---
-- - [1, 10]
- - [2, 10]
+- true
...
-s:drop()
+box.begin();
---
...
--- Narrowing indexed field type entails index rebuild.
-s = box.schema.space.create('test', {engine = 'vinyl'})
+for i = 1, 1000 do
+ if (i % 100 == 0) then
+ box.commit()
+ box.begin()
+ end
+ if i % 300 == 0 then
+ box.snapshot()
+ end
+ box.space.test:replace{i, i, i}
+end;
---
...
-_ = s:create_index('i1')
+box.commit();
---
...
-_ = s:create_index('i2', {parts = {2, 'integer'}})
+last_val = 1000;
---
...
-_ = s:create_index('i3', {parts = {{3, 'string', is_nullable = true}}})
+function gen_load()
+ fiber.sleep(0.001)
+ local s = box.space.test
+ for i = 1, 200 do
+ local op = math.random(4)
+ local key = math.random(1000)
+ local val1 = math.random(1000)
+ local val2 = last_val + 1
+ last_val = val2
+ if op == 1 then
+ pcall(s.insert, s, {key, val1, val2})
+ elseif op == 2 then
+ pcall(s.replace, s, {key, val1, val2})
+ elseif op == 3 then
+ pcall(s.delete, s, {key})
+ elseif op == 4 then
+ pcall(s.upsert, s, {key, val1, val2}, {{'=', 2, val1}, {'=', 3, val2}})
+ end
+ end
+end;
---
...
-_ = s:replace{1, 1, 'test'}
+test_run:cmd("setopt delimiter ''");
---
+- true
...
--- Should fail with 'Vinyl does not support building an index for a non-empty space'.
-s.index.i2:alter{parts = {2, 'unsigned'}}
+ch = fiber.channel(1)
---
-- error: Vinyl does not support building an index for a non-empty space
...
-s.index.i3:alter{parts = {{3, 'string', is_nullable = false}}}
+_ = fiber.create(function() gen_load() ch:put(true) end)
---
-- error: Vinyl does not support building an index for a non-empty space
...
-s:drop()
+_ = box.space.test:create_index('sk', {unique = false, parts = {2, 'unsigned'}})
---
...
+ch:get()
+---
+- true
+...
+_ = fiber.create(function() gen_load() ch:put(true) end)
+---
+...
+_ = box.space.test:create_index('tk', {unique = true, parts = {3, 'unsigned'}})
+---
+...
+ch:get()
+---
+- true
+...
+box.space.test.index.pk:count() == box.space.test.index.sk:count()
+---
+- true
+...
+box.space.test.index.pk:count() == box.space.test.index.tk:count()
+---
+- true
+...
+test_run:cmd("restart server default")
+box.space.test.index.pk:count() == box.space.test.index.sk:count()
+---
+- true
+...
+box.space.test.index.pk:count() == box.space.test.index.tk:count()
+---
+- true
+...
+box.snapshot()
+---
+- ok
+...
+box.space.test.index.pk:count() == box.space.test.index.sk:count()
+---
+- true
+...
+box.space.test.index.pk:count() == box.space.test.index.tk:count()
+---
+- true
+...
+box.space.test:drop()
+---
+...
+--
+-- Check that creation of a secondary index triggers dump
+-- if memory quota is exceeded.
+--
+test_run:cmd("create server test with script='vinyl/low_quota.lua'")
+---
+- true
+...
+test_run:cmd("start server test with args='1048576'")
+---
+- true
+...
+test_run:cmd("switch test")
+---
+- true
+...
+_ = box.schema.space.create('test', {engine = 'vinyl'})
+---
+...
+_ = box.space.test:create_index('pk')
+---
+...
+pad = string.rep('x', 1000)
+---
+...
+test_run:cmd("setopt delimiter ';'")
+---
+- true
+...
+box.begin();
+---
+...
+for i = 1, 1000 do
+ if (i % 100 == 0) then
+ box.commit()
+ box.begin()
+ end
+ box.space.test:replace{i, i, pad}
+end;
+---
+...
+box.commit();
+---
+...
+test_run:cmd("setopt delimiter ''");
+---
+- true
+...
+_ = box.space.test:create_index('sk', {parts = {2, 'unsigned', 3, 'string'}})
+---
+...
+box.space.test.index.sk:stat().disk.dump.count > 1
+---
+- true
+...
+box.space.test.index.sk:count()
+---
+- 1000
+...
+test_run:cmd("restart server test with args='1048576'")
+box.space.test.index.sk:count()
+---
+- 1000
+...
+box.space.test.index.sk:drop()
+---
+...
+box.snapshot()
+---
+- ok
+...
+--
+-- Check that run files left from an index we failed to build
+-- are removed by garbage collection.
+--
+fio = require('fio')
+---
+...
+box.cfg{checkpoint_count = 1}
+---
+...
+_ = box.space.test:replace{1001, 1, string.rep('x', 1000)}
+---
+...
+box.space.test:create_index('sk', {parts = {2, 'unsigned', 3, 'string'}})
+---
+- error: Duplicate key exists in unique index 'sk' in space 'test'
+...
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, box.space.test.id, 1)) > 0
+---
+- true
+...
+box.snapshot()
+---
+- ok
+...
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, box.space.test.id, 1)) == 0
+---
+- true
+...
+box.space.test:drop()
+---
+...
+test_run:cmd("switch default")
+---
+- true
+...
+test_run:cmd("stop server test")
+---
+- true
+...
+test_run:cmd("cleanup server test")
+---
+- true
+...
diff --git a/test/vinyl/ddl.test.lua b/test/vinyl/ddl.test.lua
index f1401df4..95dd5a11 100644
--- a/test/vinyl/ddl.test.lua
+++ b/test/vinyl/ddl.test.lua
@@ -23,76 +23,23 @@ space = box.schema.space.create('test', { engine = 'vinyl' })
index = space:create_index('primary', {type = 'hash'})
space:drop()
--- creation of a new index and altering the definition of an existing
--- index are unsupported for non-empty spaces
+-- rebuild of the primary index is not supported
space = box.schema.space.create('test', { engine = 'vinyl' })
-index = space:create_index('primary')
-space:insert({1})
--- fail because of wrong tuple format {1}, but need {1, ...}
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
-#box.space._index:select({space.id})
-box.space._index:get{space.id, 0}[6]
-space:drop()
-
-space = box.schema.space.create('test', { engine = 'vinyl' })
-index = space:create_index('primary')
-space:insert({1, 2})
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
-#box.space._index:select({space.id})
-box.space._index:get{space.id, 0}[6]
-space:drop()
-
-space = box.schema.space.create('test', { engine = 'vinyl' })
-index = space:create_index('primary')
-space:insert({1, 2})
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
-#box.space._index:select({space.id})
-box.space._index:get{space.id, 0}[6]
-space:delete({1})
-
--- must fail because vy_mems have data
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
-box.snapshot()
-while space.index.primary:stat().rows ~= 0 do fiber.sleep(0.01) end
-
--- after a dump REPLACE + DELETE = nothing, so the space is empty now and
--- can be altered.
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
-#box.space._index:select({space.id})
-box.space._index:get{space.id, 0}[6]
-space:insert({1, 2})
-index:select{}
-index2:select{}
-space:drop()
-
-space = box.schema.space.create('test', { engine = 'vinyl' })
-index = space:create_index('primary', { run_count_per_level = 2 })
-space:insert({1, 2})
-box.snapshot()
-space:delete({1})
+pk = space:create_index('pk', {run_count_per_level = 1, run_size_ratio = 10})
+space:replace{1, 1}
+pk:alter{parts = {2, 'unsigned'}} -- error: mem not empty
box.snapshot()
-while space.index.primary:stat().run_count ~= 2 do fiber.sleep(0.01) end
--- must fail because vy_runs have data
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
-
--- After compaction the REPLACE + DELETE + DELETE = nothing, so
--- the space is now empty and can be altered.
-space:delete({1})
--- Make sure the run is big enough to trigger compaction.
-space:replace({2, 3})
-space:delete({2})
+pk:alter{parts = {2, 'unsigned'}} -- error: run not empty
+space:replace{2, 2}
+space:delete{1}
+space:delete{2}
+pk:alter{parts = {2, 'unsigned'}} -- error: mem/run not empty
box.snapshot()
--- Wait until the dump is finished.
-while space.index.primary:stat().rows ~= 0 do fiber.sleep(0.01) end
-index2 = space:create_index('secondary', { parts = {2, 'unsigned'} })
-space.index.primary:alter({parts = {1, 'unsigned', 2, 'unsigned'}})
-
+-- wait for compaction to complete
+while pk:stat().disk.compact.count == 0 do fiber.sleep(0.01) end
+pk:alter{parts = {2, 'unsigned'}} -- success: space is empty now
+space:replace{1, 2}
+space:get(2)
space:drop()
--
@@ -266,26 +213,136 @@ box.space._index:insert{512, 0, 'pk', 'tree', {unique = true}, {{0, 'unsigned'}}
box.space.test.index.pk
box.space.test:drop()
--- gh-2449 change 'unique' index property from true to false
-s = box.schema.space.create('test', { engine = 'vinyl' })
-_ = s:create_index('primary')
-_ = s:create_index('secondary', {unique = true, parts = {2, 'unsigned'}})
-s:insert{1, 10}
-s.index.secondary:alter{unique = false} -- ok
-s.index.secondary.unique
-s.index.secondary:alter{unique = true} -- error
-s.index.secondary.unique
-s:insert{2, 10}
-s.index.secondary:select(10)
-s:drop()
-
-- Narrowing indexed field type entails index rebuild.
s = box.schema.space.create('test', {engine = 'vinyl'})
-_ = s:create_index('i1')
-_ = s:create_index('i2', {parts = {2, 'integer'}})
-_ = s:create_index('i3', {parts = {{3, 'string', is_nullable = true}}})
-_ = s:replace{1, 1, 'test'}
--- Should fail with 'Vinyl does not support building an index for a non-empty space'.
-s.index.i2:alter{parts = {2, 'unsigned'}}
-s.index.i3:alter{parts = {{3, 'string', is_nullable = false}}}
+pk = s:create_index('pk', {parts = {1, 'unsigned'}})
+s:replace{1}
+-- Extending field type is allowed without rebuild.
+pk:alter{parts = {1, 'integer'}}
+-- Should fail as we do not support rebuilding the primary index of a non-empty space.
+pk:alter{parts = {1, 'unsigned'}}
+s:replace{-1}
s:drop()
+
+--
+-- Check that all modifications done to the space during index build
+-- are reflected in the new index.
+--
+math.randomseed(os.time())
+
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk')
+
+test_run:cmd("setopt delimiter ';'")
+
+box.begin();
+for i = 1, 1000 do
+ if (i % 100 == 0) then
+ box.commit()
+ box.begin()
+ end
+ if i % 300 == 0 then
+ box.snapshot()
+ end
+ box.space.test:replace{i, i, i}
+end;
+box.commit();
+
+last_val = 1000;
+
+function gen_load()
+ fiber.sleep(0.001)
+ local s = box.space.test
+ for i = 1, 200 do
+ local op = math.random(4)
+ local key = math.random(1000)
+ local val1 = math.random(1000)
+ local val2 = last_val + 1
+ last_val = val2
+ if op == 1 then
+ pcall(s.insert, s, {key, val1, val2})
+ elseif op == 2 then
+ pcall(s.replace, s, {key, val1, val2})
+ elseif op == 3 then
+ pcall(s.delete, s, {key})
+ elseif op == 4 then
+ pcall(s.upsert, s, {key, val1, val2}, {{'=', 2, val1}, {'=', 3, val2}})
+ end
+ end
+end;
+
+test_run:cmd("setopt delimiter ''");
+
+ch = fiber.channel(1)
+
+_ = fiber.create(function() gen_load() ch:put(true) end)
+_ = box.space.test:create_index('sk', {unique = false, parts = {2, 'unsigned'}})
+ch:get()
+
+_ = fiber.create(function() gen_load() ch:put(true) end)
+_ = box.space.test:create_index('tk', {unique = true, parts = {3, 'unsigned'}})
+ch:get()
+
+box.space.test.index.pk:count() == box.space.test.index.sk:count()
+box.space.test.index.pk:count() == box.space.test.index.tk:count()
+
+test_run:cmd("restart server default")
+
+box.space.test.index.pk:count() == box.space.test.index.sk:count()
+box.space.test.index.pk:count() == box.space.test.index.tk:count()
+box.snapshot()
+box.space.test.index.pk:count() == box.space.test.index.sk:count()
+box.space.test.index.pk:count() == box.space.test.index.tk:count()
+box.space.test:drop()
+
+--
+-- Check that creation of a secondary index triggers dump
+-- if memory quota is exceeded.
+--
+test_run:cmd("create server test with script='vinyl/low_quota.lua'")
+test_run:cmd("start server test with args='1048576'")
+test_run:cmd("switch test")
+
+_ = box.schema.space.create('test', {engine = 'vinyl'})
+_ = box.space.test:create_index('pk')
+
+pad = string.rep('x', 1000)
+
+test_run:cmd("setopt delimiter ';'")
+box.begin();
+for i = 1, 1000 do
+ if (i % 100 == 0) then
+ box.commit()
+ box.begin()
+ end
+ box.space.test:replace{i, i, pad}
+end;
+box.commit();
+test_run:cmd("setopt delimiter ''");
+
+_ = box.space.test:create_index('sk', {parts = {2, 'unsigned', 3, 'string'}})
+box.space.test.index.sk:stat().disk.dump.count > 1
+box.space.test.index.sk:count()
+
+test_run:cmd("restart server test with args='1048576'")
+
+box.space.test.index.sk:count()
+box.space.test.index.sk:drop()
+box.snapshot()
+
+--
+-- Check that run files left from an index we failed to build
+-- are removed by garbage collection.
+--
+fio = require('fio')
+box.cfg{checkpoint_count = 1}
+_ = box.space.test:replace{1001, 1, string.rep('x', 1000)}
+box.space.test:create_index('sk', {parts = {2, 'unsigned', 3, 'string'}})
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, box.space.test.id, 1)) > 0
+box.snapshot()
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, box.space.test.id, 1)) == 0
+box.space.test:drop()
+
+test_run:cmd("switch default")
+test_run:cmd("stop server test")
+test_run:cmd("cleanup server test")
diff --git a/test/vinyl/errinj.result b/test/vinyl/errinj.result
index 983acc20..28271fc9 100644
--- a/test/vinyl/errinj.result
+++ b/test/vinyl/errinj.result
@@ -1532,3 +1532,315 @@ box.backup.stop()
s:drop()
---
...
+--
+-- gh-2449: change 'unique' index property from true to false
+-- is done without index rebuild.
+--
+s = box.schema.space.create('test', { engine = 'vinyl' })
+---
+...
+_ = s:create_index('primary')
+---
+...
+_ = s:create_index('secondary', {unique = true, parts = {2, 'unsigned'}})
+---
+...
+s:insert{1, 10}
+---
+- [1, 10]
+...
+box.snapshot()
+---
+- ok
+...
+errinj.set("ERRINJ_VY_READ_PAGE", true);
+---
+- ok
+...
+s.index.secondary:alter{unique = false} -- ok
+---
+...
+s.index.secondary.unique
+---
+- false
+...
+s.index.secondary:alter{unique = true} -- error
+---
+- error: Error injection 'vinyl page read'
+...
+s.index.secondary.unique
+---
+- false
+...
+errinj.set("ERRINJ_VY_READ_PAGE", false);
+---
+- ok
+...
+s:insert{2, 10}
+---
+- [2, 10]
+...
+s.index.secondary:select(10)
+---
+- - [1, 10]
+ - [2, 10]
+...
+s:drop()
+---
+...
+--
+-- Check that ALTER is aborted if a tuple inserted during index build
+-- doesn't conform to the new format.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+---
+...
+_ = s:create_index('pk', {page_size = 16})
+---
+...
+pad = string.rep('x', 16)
+---
+...
+for i = 101, 200 do s:replace{i, i, pad} end
+---
+...
+box.snapshot()
+---
+- ok
+...
+ch = fiber.channel(1)
+---
+...
+test_run:cmd("setopt delimiter ';'")
+---
+- true
+...
+_ = fiber.create(function()
+ fiber.sleep(0.01)
+ for i = 1, 100 do
+ s:replace{i}
+ end
+ ch:put(true)
+end);
+---
+...
+test_run:cmd("setopt delimiter ''");
+---
+- true
+...
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0.001)
+---
+- ok
+...
+s:create_index('sk', {parts = {2, 'unsigned'}}) -- must fail
+---
+- error: Tuple field count 1 is less than required by space format or defined indexes
+ (expected at least 2)
+...
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0)
+---
+- ok
+...
+ch:get()
+---
+- true
+...
+s:count() -- 200
+---
+- 200
+...
+s:drop()
+---
+...
+--
+-- Check that ALTER is aborted if a tuple inserted during index build
+-- violates unique constraint.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+---
+...
+_ = s:create_index('pk', {page_size = 16})
+---
+...
+pad = string.rep('x', 16)
+---
+...
+for i = 101, 200 do s:replace{i, i, pad} end
+---
+...
+box.snapshot()
+---
+- ok
+...
+ch = fiber.channel(1)
+---
+...
+test_run:cmd("setopt delimiter ';'")
+---
+- true
+...
+_ = fiber.create(function()
+ fiber.sleep(0.01)
+ for i = 1, 100 do
+ s:replace{i, i + 1}
+ end
+ ch:put(true)
+end);
+---
+...
+test_run:cmd("setopt delimiter ''");
+---
+- true
+...
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0.001)
+---
+- ok
+...
+s:create_index('sk', {parts = {2, 'unsigned'}}) -- must fail
+---
+- error: Duplicate key exists in unique index 'sk' in space 'test'
+...
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0)
+---
+- ok
+...
+ch:get()
+---
+- true
+...
+s:count() -- 200
+---
+- 200
+...
+s:drop()
+---
+...
+--
+-- Check that modifications done to the space during the final dump
+-- of a newly built index are recovered properly.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+---
+...
+_ = s:create_index('pk')
+---
+...
+for i = 1, 5 do s:replace{i, i} end
+---
+...
+errinj.set("ERRINJ_VY_RUN_WRITE_TIMEOUT", 0.1)
+---
+- ok
+...
+ch = fiber.channel(1)
+---
+...
+_ = fiber.create(function() s:create_index('sk', {parts = {2, 'integer'}}) ch:put(true) end)
+---
+...
+errinj.set("ERRINJ_VY_RUN_WRITE_TIMEOUT", 0)
+---
+- ok
+...
+fiber.sleep(0.01)
+---
+...
+_ = s:delete{1}
+---
+...
+_ = s:replace{2, -2}
+---
+...
+_ = s:delete{2}
+---
+...
+_ = s:replace{3, -3}
+---
+...
+_ = s:replace{3, -2}
+---
+...
+_ = s:replace{3, -1}
+---
+...
+_ = s:delete{3}
+---
+...
+_ = s:upsert({3, 3}, {{'=', 2, 1}})
+---
+...
+_ = s:upsert({3, 3}, {{'=', 2, 2}})
+---
+...
+_ = s:delete{3}
+---
+...
+_ = s:replace{4, -1}
+---
+...
+_ = s:replace{4, -2}
+---
+...
+_ = s:replace{4, -4}
+---
+...
+_ = s:upsert({5, 1}, {{'=', 2, 1}})
+---
+...
+_ = s:upsert({5, 2}, {{'=', 2, -5}})
+---
+...
+_ = s:replace{6, -6}
+---
+...
+_ = s:upsert({7, -7}, {{'=', 2, -7}})
+---
+...
+ch:get()
+---
+- true
+...
+s.index.sk:select()
+---
+- - [7, -7]
+ - [6, -6]
+ - [5, -5]
+ - [4, -4]
+...
+s.index.sk:stat().memory.rows
+---
+- 27
+...
+test_run:cmd('restart server default')
+s = box.space.test
+---
+...
+s.index.sk:select()
+---
+- - [7, -7]
+ - [6, -6]
+ - [5, -5]
+ - [4, -4]
+...
+s.index.sk:stat().memory.rows
+---
+- 27
+...
+box.snapshot()
+---
+- ok
+...
+s.index.sk:select()
+---
+- - [7, -7]
+ - [6, -6]
+ - [5, -5]
+ - [4, -4]
+...
+s.index.sk:stat().memory.rows
+---
+- 0
+...
+s:drop()
+---
+...
diff --git a/test/vinyl/errinj.test.lua b/test/vinyl/errinj.test.lua
index 6b2f4c4c..000067d3 100644
--- a/test/vinyl/errinj.test.lua
+++ b/test/vinyl/errinj.test.lua
@@ -602,3 +602,137 @@ for _, f in pairs(files) do if not fio.path.exists(f) then table.insert(missing,
missing
box.backup.stop()
s:drop()
+
+--
+-- gh-2449: change 'unique' index property from true to false
+-- is done without index rebuild.
+--
+s = box.schema.space.create('test', { engine = 'vinyl' })
+_ = s:create_index('primary')
+_ = s:create_index('secondary', {unique = true, parts = {2, 'unsigned'}})
+s:insert{1, 10}
+box.snapshot()
+errinj.set("ERRINJ_VY_READ_PAGE", true);
+s.index.secondary:alter{unique = false} -- ok
+s.index.secondary.unique
+s.index.secondary:alter{unique = true} -- error
+s.index.secondary.unique
+errinj.set("ERRINJ_VY_READ_PAGE", false);
+s:insert{2, 10}
+s.index.secondary:select(10)
+s:drop()
+
+--
+-- Check that ALTER is aborted if a tuple inserted during index build
+-- doesn't conform to the new format.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk', {page_size = 16})
+
+pad = string.rep('x', 16)
+for i = 101, 200 do s:replace{i, i, pad} end
+box.snapshot()
+
+ch = fiber.channel(1)
+test_run:cmd("setopt delimiter ';'")
+_ = fiber.create(function()
+ fiber.sleep(0.01)
+ for i = 1, 100 do
+ s:replace{i}
+ end
+ ch:put(true)
+end);
+test_run:cmd("setopt delimiter ''");
+
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0.001)
+s:create_index('sk', {parts = {2, 'unsigned'}}) -- must fail
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0)
+
+ch:get()
+
+s:count() -- 200
+s:drop()
+
+--
+-- Check that ALTER is aborted if a tuple inserted during index build
+-- violates unique constraint.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk', {page_size = 16})
+
+pad = string.rep('x', 16)
+for i = 101, 200 do s:replace{i, i, pad} end
+box.snapshot()
+
+ch = fiber.channel(1)
+test_run:cmd("setopt delimiter ';'")
+_ = fiber.create(function()
+ fiber.sleep(0.01)
+ for i = 1, 100 do
+ s:replace{i, i + 1}
+ end
+ ch:put(true)
+end);
+test_run:cmd("setopt delimiter ''");
+
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0.001)
+s:create_index('sk', {parts = {2, 'unsigned'}}) -- must fail
+errinj.set("ERRINJ_VY_READ_PAGE_TIMEOUT", 0)
+
+ch:get()
+
+s:count() -- 200
+s:drop()
+
+--
+-- Check that modifications done to the space during the final dump
+-- of a newly built index are recovered properly.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk')
+
+for i = 1, 5 do s:replace{i, i} end
+
+errinj.set("ERRINJ_VY_RUN_WRITE_TIMEOUT", 0.1)
+ch = fiber.channel(1)
+_ = fiber.create(function() s:create_index('sk', {parts = {2, 'integer'}}) ch:put(true) end)
+errinj.set("ERRINJ_VY_RUN_WRITE_TIMEOUT", 0)
+
+fiber.sleep(0.01)
+
+_ = s:delete{1}
+_ = s:replace{2, -2}
+_ = s:delete{2}
+_ = s:replace{3, -3}
+_ = s:replace{3, -2}
+_ = s:replace{3, -1}
+_ = s:delete{3}
+_ = s:upsert({3, 3}, {{'=', 2, 1}})
+_ = s:upsert({3, 3}, {{'=', 2, 2}})
+_ = s:delete{3}
+_ = s:replace{4, -1}
+_ = s:replace{4, -2}
+_ = s:replace{4, -4}
+_ = s:upsert({5, 1}, {{'=', 2, 1}})
+_ = s:upsert({5, 2}, {{'=', 2, -5}})
+_ = s:replace{6, -6}
+_ = s:upsert({7, -7}, {{'=', 2, -7}})
+
+ch:get()
+
+s.index.sk:select()
+s.index.sk:stat().memory.rows
+
+test_run:cmd('restart server default')
+
+s = box.space.test
+
+s.index.sk:select()
+s.index.sk:stat().memory.rows
+
+box.snapshot()
+
+s.index.sk:select()
+s.index.sk:stat().memory.rows
+
+s:drop()
diff --git a/test/vinyl/errinj_gc.result b/test/vinyl/errinj_gc.result
index 0cbca654..7c963103 100644
--- a/test/vinyl/errinj_gc.result
+++ b/test/vinyl/errinj_gc.result
@@ -180,9 +180,6 @@ s:select()
- - [100, '12345']
- [200, '67890']
...
---
--- Cleanup.
---
s:drop()
---
...
@@ -199,3 +196,78 @@ temp:drop()
box.cfg{checkpoint_count = default_checkpoint_count}
---
...
+--
+-- Check that if we failed to clean up an incomplete index before restart,
+-- we will clean it up after recovery.
+--
+fio = require('fio')
+---
+...
+fiber = require('fiber')
+---
+...
+s = box.schema.space.create('test', {engine = 'vinyl'})
+---
+...
+_ = s:create_index('pk')
+---
+...
+_ = s:insert{1, 1}
+---
+...
+box.error.injection.set('ERRINJ_WAL_DELAY', true)
+---
+- ok
+...
+ch = fiber.channel(1)
+---
+...
+_ = fiber.create(function() pcall(s.create_index, s, 'sk', {parts = {2, 'unsigned'}}) ch:put(true) end)
+---
+...
+-- wait for ALTER to stall on WAL after preparing the new index
+while s.index.pk:stat().disk.dump.count == 0 do fiber.sleep(0.001) end
+---
+...
+box.error.injection.set('ERRINJ_VY_GC', true)
+---
+- ok
+...
+box.error.injection.set('ERRINJ_VY_LOG_FLUSH', true);
+---
+- ok
+...
+box.error.injection.set("ERRINJ_WAL_WRITE", true)
+---
+- ok
+...
+box.error.injection.set('ERRINJ_WAL_DELAY', false)
+---
+- ok
+...
+ch:get()
+---
+- true
+...
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, s.id, 1)) > 0
+---
+- true
+...
+test_run:cmd('restart server default')
+box.snapshot()
+---
+- ok
+...
+fio = require('fio')
+---
+...
+s = box.space.test
+---
+...
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, s.id, 1)) == 0
+---
+- true
+...
+s:drop()
+---
+...
diff --git a/test/vinyl/errinj_gc.test.lua b/test/vinyl/errinj_gc.test.lua
index ee590a4c..4c3d1bb3 100644
--- a/test/vinyl/errinj_gc.test.lua
+++ b/test/vinyl/errinj_gc.test.lua
@@ -85,10 +85,6 @@ file_count()
s:select()
---
--- Cleanup.
---
-
s:drop()
gc()
file_count()
@@ -96,3 +92,37 @@ file_count()
temp:drop()
box.cfg{checkpoint_count = default_checkpoint_count}
+
+--
+-- Check that if we failed to clean up an incomplete index before restart,
+-- we will clean it up after recovery.
+--
+fio = require('fio')
+fiber = require('fiber')
+
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk')
+_ = s:insert{1, 1}
+
+box.error.injection.set('ERRINJ_WAL_DELAY', true)
+ch = fiber.channel(1)
+_ = fiber.create(function() pcall(s.create_index, s, 'sk', {parts = {2, 'unsigned'}}) ch:put(true) end)
+
+-- wait for ALTER to stall on WAL after preparing the new index
+while s.index.pk:stat().disk.dump.count == 0 do fiber.sleep(0.001) end
+
+box.error.injection.set('ERRINJ_VY_GC', true)
+box.error.injection.set('ERRINJ_VY_LOG_FLUSH', true);
+box.error.injection.set("ERRINJ_WAL_WRITE", true)
+box.error.injection.set('ERRINJ_WAL_DELAY', false)
+ch:get()
+
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, s.id, 1)) > 0
+
+test_run:cmd('restart server default')
+box.snapshot()
+
+fio = require('fio')
+s = box.space.test
+#fio.listdir(fio.pathjoin(box.cfg.vinyl_dir, s.id, 1)) == 0
+s:drop()
diff --git a/test/vinyl/errinj_vylog.result b/test/vinyl/errinj_vylog.result
index ca23cb45..9ced03df 100644
--- a/test/vinyl/errinj_vylog.result
+++ b/test/vinyl/errinj_vylog.result
@@ -276,3 +276,60 @@ s1:drop()
s2:drop()
---
...
+--
+-- Check that an index that was prepared, but not committed,
+-- is recovered properly.
+--
+fiber = require('fiber')
+---
+...
+s = box.schema.space.create('test', {engine = 'vinyl'})
+---
+...
+_ = s:create_index('pk')
+---
+...
+_ = s:insert{1, 1}
+---
+...
+box.error.injection.set('ERRINJ_WAL_DELAY', true)
+---
+- ok
+...
+ch = fiber.channel(1)
+---
+...
+_ = fiber.create(function() s:create_index('sk', {parts = {2, 'unsigned'}}) ch:put(true) end)
+---
+...
+-- wait for ALTER to stall on WAL after preparing the new index
+while s.index.pk:stat().disk.dump.count == 0 do fiber.sleep(0.001) end
+---
+...
+box.error.injection.set('ERRINJ_VY_LOG_FLUSH', true);
+---
+- ok
+...
+box.error.injection.set('ERRINJ_WAL_DELAY', false)
+---
+- ok
+...
+ch:get()
+---
+- true
+...
+test_run:cmd('restart server default')
+s = box.space.test
+---
+...
+s.index.pk:select()
+---
+- - [1, 1]
+...
+s.index.sk:select()
+---
+- - [1, 1]
+...
+s:drop()
+---
+...
diff --git a/test/vinyl/errinj_vylog.test.lua b/test/vinyl/errinj_vylog.test.lua
index 3d90755d..c1fd517d 100644
--- a/test/vinyl/errinj_vylog.test.lua
+++ b/test/vinyl/errinj_vylog.test.lua
@@ -130,3 +130,32 @@ s1:select()
s2:select()
s1:drop()
s2:drop()
+
+--
+-- Check that an index that was prepared, but not committed,
+-- is recovered properly.
+--
+fiber = require('fiber')
+
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk')
+_ = s:insert{1, 1}
+
+box.error.injection.set('ERRINJ_WAL_DELAY', true)
+ch = fiber.channel(1)
+_ = fiber.create(function() s:create_index('sk', {parts = {2, 'unsigned'}}) ch:put(true) end)
+
+-- wait for ALTER to stall on WAL after preparing the new index
+while s.index.pk:stat().disk.dump.count == 0 do fiber.sleep(0.001) end
+
+box.error.injection.set('ERRINJ_VY_LOG_FLUSH', true);
+box.error.injection.set('ERRINJ_WAL_DELAY', false)
+ch:get()
+
+test_run:cmd('restart server default')
+
+s = box.space.test
+s.index.pk:select()
+s.index.sk:select()
+
+s:drop()
diff --git a/test/vinyl/gh.result b/test/vinyl/gh.result
index 47695acb..76beab09 100644
--- a/test/vinyl/gh.result
+++ b/test/vinyl/gh.result
@@ -144,7 +144,7 @@ s:insert{5, 5}
...
s.index.primary:alter({parts={2,'unsigned'}})
---
-- error: Vinyl does not support building an index for a non-empty space
+- error: Vinyl does not support rebuilding the primary index of a non-empty space
...
s:drop()
---
--
2.11.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v3 0/3] Allow to build indexes for vinyl spaces
2018-06-07 10:56 [PATCH v3 0/3] Allow to build indexes for vinyl spaces Vladimir Davydov
` (2 preceding siblings ...)
2018-06-07 10:56 ` [PATCH v3 3/3] vinyl: allow to build secondary index for non-empty space Vladimir Davydov
@ 2018-06-07 14:02 ` Konstantin Osipov
3 siblings, 0 replies; 5+ messages in thread
From: Konstantin Osipov @ 2018-06-07 14:02 UTC (permalink / raw)
To: Vladimir Davydov; +Cc: tarantool-patches
* Vladimir Davydov <vdavydov.dev@gmail.com> [18/06/07 14:27]:
I pushed this patch set.
Vlad, if you have any comment re the last patch in the stack,
please write them here.
I believe we need to replace region_alloc with a new allocator
rather than wasting time on each individual case when copying
memory could be avoided.
Let's see how the community likes this feature.
> This patch set implements the ability to build secondary indexes for
> non-empty vinyl spaces. For implementation details, see patch 3.
>
> https://github.com/tarantool/tarantool/issues/1653
> https://github.com/tarantool/tarantool/commits/vy-allow-to-build-secondary-indexes
>
> - When building a new index, use original tuple LSNs instead
> of max LSN to the time.
This is a very good fix.
We only need to get rid of vylog to make the whole thing perfect,
and now we are a little closer to knowing how to do it.
--
Konstantin Osipov, Moscow, Russia, +7 903 626 22 32
http://tarantool.io - www.twitter.com/kostja_osipov
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2018-06-07 14:02 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-06-07 10:56 [PATCH v3 0/3] Allow to build indexes for vinyl spaces Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 1/3] vinyl: do not yield on dump completion Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 2/3] vinyl: relax limitation imposed on run min/max lsn Vladimir Davydov
2018-06-07 10:56 ` [PATCH v3 3/3] vinyl: allow to build secondary index for non-empty space Vladimir Davydov
2018-06-07 14:02 ` [PATCH v3 0/3] Allow to build indexes for vinyl spaces Konstantin Osipov
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox